File size: 91,877 Bytes
b381930
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
# Copyright 2021 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import time
import unittest

import numpy as np
import pytest
from datasets import Audio, load_dataset
from huggingface_hub import AutomaticSpeechRecognitionOutput, hf_hub_download, snapshot_download

from transformers import (
    MODEL_FOR_CTC_MAPPING,
    MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING,
    AutoFeatureExtractor,
    AutoModelForCausalLM,
    AutoModelForSpeechSeq2Seq,
    AutoProcessor,
    AutoTokenizer,
    Speech2TextForConditionalGeneration,
    Wav2Vec2ForCTC,
    WhisperForConditionalGeneration,
)
from transformers.pipelines import AutomaticSpeechRecognitionPipeline, pipeline
from transformers.pipelines.audio_utils import chunk_bytes_iter, ffmpeg_microphone_live
from transformers.pipelines.automatic_speech_recognition import _find_timestamp_sequence, chunk_iter
from transformers.testing_utils import (
    compare_pipeline_output_to_hub_spec,
    is_pipeline_test,
    is_torch_available,
    nested_simplify,
    require_pyctcdecode,
    require_tf,
    require_torch,
    require_torch_accelerator,
    require_torchaudio,
    slow,
    torch_device,
)

from .test_pipelines_common import ANY


if is_torch_available():
    import torch


# We can't use this mixin because it assumes TF support.
# from .test_pipelines_common import CustomInputPipelineCommonMixin


@is_pipeline_test
class AutomaticSpeechRecognitionPipelineTests(unittest.TestCase):
    model_mapping = dict(
        (list(MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING.items()) if MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING else [])
        + (MODEL_FOR_CTC_MAPPING.items() if MODEL_FOR_CTC_MAPPING else [])
    )

    def get_test_pipeline(
        self,
        model,
        tokenizer=None,
        image_processor=None,
        feature_extractor=None,
        processor=None,
        torch_dtype="float32",
    ):
        if tokenizer is None:
            # Side effect of no Fast Tokenizer class for these model, so skipping
            # But the slow tokenizer test should still run as they're quite small
            self.skipTest(reason="No tokenizer available")

        speech_recognizer = AutomaticSpeechRecognitionPipeline(
            model=model,
            tokenizer=tokenizer,
            feature_extractor=feature_extractor,
            image_processor=image_processor,
            processor=processor,
            torch_dtype=torch_dtype,
        )

        # test with a raw waveform
        audio = np.zeros((34000,))
        audio2 = np.zeros((14000,))
        return speech_recognizer, [audio, audio2]

    def run_pipeline_test(self, speech_recognizer, examples):
        audio = np.zeros((34000,))
        outputs = speech_recognizer(audio)
        self.assertEqual(outputs, {"text": ANY(str)})

        compare_pipeline_output_to_hub_spec(outputs, AutomaticSpeechRecognitionOutput)

        # Striding
        audio = {"raw": audio, "stride": (0, 4000), "sampling_rate": speech_recognizer.feature_extractor.sampling_rate}
        if speech_recognizer.type == "ctc":
            outputs = speech_recognizer(audio)
            self.assertEqual(outputs, {"text": ANY(str)})
        elif "Whisper" in speech_recognizer.model.__class__.__name__:
            outputs = speech_recognizer(audio)
            self.assertEqual(outputs, {"text": ANY(str)})
        else:
            # Non CTC models cannot use striding.
            with self.assertRaises(ValueError):
                outputs = speech_recognizer(audio)

        # Timestamps
        audio = np.zeros((34000,))
        if speech_recognizer.type == "ctc":
            outputs = speech_recognizer(audio, return_timestamps="char")
            self.assertIsInstance(outputs["chunks"], list)
            n = len(outputs["chunks"])
            self.assertEqual(
                outputs,
                {
                    "text": ANY(str),
                    "chunks": [{"text": ANY(str), "timestamp": (ANY(float), ANY(float))} for i in range(n)],
                },
            )

            outputs = speech_recognizer(audio, return_timestamps="word")
            self.assertIsInstance(outputs["chunks"], list)
            n = len(outputs["chunks"])
            self.assertEqual(
                outputs,
                {
                    "text": ANY(str),
                    "chunks": [{"text": ANY(str), "timestamp": (ANY(float), ANY(float))} for i in range(n)],
                },
            )
        elif "Whisper" in speech_recognizer.model.__class__.__name__:
            outputs = speech_recognizer(audio, return_timestamps=True)
            self.assertIsInstance(outputs["chunks"], list)
            nb_chunks = len(outputs["chunks"])
            self.assertGreater(nb_chunks, 0)
            self.assertEqual(
                outputs,
                {
                    "text": ANY(str),
                    "chunks": [{"text": ANY(str), "timestamp": (ANY(float), ANY(float))} for i in range(nb_chunks)],
                },
            )
        else:
            # Non CTC models cannot use return_timestamps
            with self.assertRaisesRegex(
                ValueError, "^We cannot return_timestamps yet on non-CTC models apart from Whisper!$"
            ):
                outputs = speech_recognizer(audio, return_timestamps="char")

    @require_torch
    @slow
    def test_pt_defaults(self):
        pipeline("automatic-speech-recognition", framework="pt")

    @require_torch
    def test_small_model_pt(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="facebook/s2t-small-mustc-en-fr-st",
            tokenizer="facebook/s2t-small-mustc-en-fr-st",
            framework="pt",
        )
        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)
        output = speech_recognizer(waveform)
        self.assertEqual(output, {"text": "(Applaudissements)"})
        output = speech_recognizer(waveform, chunk_length_s=10)
        self.assertEqual(output, {"text": "(Applaudissements)"})

        # Non CTC models cannot use return_timestamps
        with self.assertRaisesRegex(
            ValueError, "^We cannot return_timestamps yet on non-CTC models apart from Whisper!$"
        ):
            _ = speech_recognizer(waveform, return_timestamps="char")

    @require_torch
    def test_small_model_pt_fp16(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="facebook/s2t-small-mustc-en-fr-st",
            tokenizer="facebook/s2t-small-mustc-en-fr-st",
            framework="pt",
            torch_dtype=torch.float16,
        )
        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)
        output = speech_recognizer(waveform)
        self.assertEqual(output, {"text": "(Applaudissements)"})
        output = speech_recognizer(waveform, chunk_length_s=10)
        self.assertEqual(output, {"text": "(Applaudissements)"})

        # Non CTC models cannot use return_timestamps
        with self.assertRaisesRegex(
            ValueError, "^We cannot return_timestamps yet on non-CTC models apart from Whisper!$"
        ):
            _ = speech_recognizer(waveform, return_timestamps="char")

    @require_torch
    def test_small_model_pt_bf16(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="facebook/s2t-small-mustc-en-fr-st",
            tokenizer="facebook/s2t-small-mustc-en-fr-st",
            framework="pt",
            torch_dtype=torch.bfloat16,
        )
        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)
        output = speech_recognizer(waveform)
        self.assertEqual(output, {"text": "(Applaudissements)"})
        output = speech_recognizer(waveform, chunk_length_s=10)
        self.assertEqual(output, {"text": "(Applaudissements)"})

        # Non CTC models cannot use return_timestamps
        with self.assertRaisesRegex(
            ValueError, "^We cannot return_timestamps yet on non-CTC models apart from Whisper!$"
        ):
            _ = speech_recognizer(waveform, return_timestamps="char")

    @slow
    @require_torch_accelerator
    def test_whisper_fp16(self):
        speech_recognizer = pipeline(
            model="openai/whisper-base",
            device=torch_device,
            torch_dtype=torch.float16,
        )
        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)
        speech_recognizer(waveform)

    @require_torch
    def test_small_model_pt_seq2seq(self):
        speech_recognizer = pipeline(
            model="hf-internal-testing/tiny-random-speech-encoder-decoder",
            framework="pt",
        )

        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)
        output = speech_recognizer(waveform)
        self.assertEqual(output, {"text": "あл ش 湯 清 ه ܬ া लᆨしث ल eか u w 全 u"})

    @require_torch
    def test_small_model_pt_seq2seq_gen_kwargs(self):
        speech_recognizer = pipeline(
            model="hf-internal-testing/tiny-random-speech-encoder-decoder",
            framework="pt",
        )

        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)
        output = speech_recognizer(waveform, max_new_tokens=10, generate_kwargs={"num_beams": 2})
        self.assertEqual(output, {"text": "あл † γ ت ב オ 束 泣 足"})

    @slow
    @require_torch
    @require_pyctcdecode
    def test_large_model_pt_with_lm(self):
        dataset = load_dataset("Narsil/asr_dummy", streaming=True, trust_remote_code=True)
        third_item = next(iter(dataset["test"].skip(3)))
        filename = third_item["file"]

        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="patrickvonplaten/wav2vec2-large-xlsr-53-spanish-with-lm",
            framework="pt",
        )
        self.assertEqual(speech_recognizer.type, "ctc_with_lm")

        output = speech_recognizer(filename)
        self.assertEqual(
            output,
            {"text": "y en las ramas medio sumergidas revoloteaban algunos pájaros de quimérico y legendario plumaje"},
        )

        # Override back to pure CTC
        speech_recognizer.type = "ctc"
        output = speech_recognizer(filename)
        # plumajre != plumaje
        self.assertEqual(
            output,
            {
                "text": (
                    "y en las ramas medio sumergidas revoloteaban algunos pájaros de quimérico y legendario plumajre"
                )
            },
        )

        speech_recognizer.type = "ctc_with_lm"
        # Simple test with CTC with LM, chunking + timestamps
        output = speech_recognizer(filename, chunk_length_s=2.0, return_timestamps="word")
        self.assertEqual(
            output,
            {
                "text": (
                    "y en las ramas medio sumergidas revoloteaban algunos pájaros de quimérico y legendario plumajcri"
                ),
                "chunks": [
                    {"text": "y", "timestamp": (0.52, 0.54)},
                    {"text": "en", "timestamp": (0.6, 0.68)},
                    {"text": "las", "timestamp": (0.74, 0.84)},
                    {"text": "ramas", "timestamp": (0.94, 1.24)},
                    {"text": "medio", "timestamp": (1.32, 1.52)},
                    {"text": "sumergidas", "timestamp": (1.56, 2.22)},
                    {"text": "revoloteaban", "timestamp": (2.36, 3.0)},
                    {"text": "algunos", "timestamp": (3.06, 3.38)},
                    {"text": "pájaros", "timestamp": (3.46, 3.86)},
                    {"text": "de", "timestamp": (3.92, 4.0)},
                    {"text": "quimérico", "timestamp": (4.08, 4.6)},
                    {"text": "y", "timestamp": (4.66, 4.68)},
                    {"text": "legendario", "timestamp": (4.74, 5.26)},
                    {"text": "plumajcri", "timestamp": (5.34, 5.74)},
                ],
            },
        )
        # CTC + LM models cannot use return_timestamps="char"
        with self.assertRaisesRegex(
            ValueError, "^CTC with LM can only predict word level timestamps, set `return_timestamps='word'`$"
        ):
            _ = speech_recognizer(filename, return_timestamps="char")

    @require_tf
    def test_small_model_tf(self):
        self.skipTest(reason="Tensorflow not supported yet.")

    @require_torch
    def test_torch_small_no_tokenizer_files(self):
        # test that model without tokenizer file cannot be loaded
        with pytest.raises(OSError):
            pipeline(
                task="automatic-speech-recognition",
                model="patrickvonplaten/tiny-wav2vec2-no-tokenizer",
                framework="pt",
            )

    @require_torch
    @slow
    def test_torch_large(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="facebook/wav2vec2-base-960h",
            tokenizer="facebook/wav2vec2-base-960h",
            framework="pt",
        )
        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)
        output = speech_recognizer(waveform)
        self.assertEqual(output, {"text": ""})

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = speech_recognizer(audio)
        self.assertEqual(output, {"text": "A MAN SAID TO THE UNIVERSE SIR I EXIST"})

    @require_torch
    @slow
    def test_torch_large_with_input_features(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="hf-audio/wav2vec2-bert-CV16-en",
            framework="pt",
        )
        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)
        output = speech_recognizer(waveform)
        self.assertEqual(output, {"text": ""})

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = speech_recognizer(audio)
        self.assertEqual(output, {"text": "a man said to the universe sir i exist"})

    @slow
    @require_torch
    def test_return_timestamps_in_preprocess(self):
        pipe = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-tiny",
            chunk_length_s=8,
            stride_length_s=1,
        )
        data = load_dataset("openslr/librispeech_asr", "clean", split="test", streaming=True, trust_remote_code=True)
        sample = next(iter(data))

        res = pipe(sample["audio"]["array"])
        self.assertEqual(res, {"text": " Conquered returned to its place amidst the tents."})

        res = pipe(sample["audio"]["array"], return_timestamps=True)
        self.assertEqual(
            res,
            {
                "text": " Conquered returned to its place amidst the tents.",
                "chunks": [{"timestamp": (0.0, 3.36), "text": " Conquered returned to its place amidst the tents."}],
            },
        )

        res = pipe(sample["audio"]["array"], return_timestamps="word")
        # fmt: off
        self.assertEqual(
            res,
            {
                'text': ' Conquered returned to its place amidst the tents.',
                'chunks': [
                    {'text': ' Conquered', 'timestamp': (0.5, 1.2)},
                    {'text': ' returned', 'timestamp': (1.2, 1.64)},
                    {'text': ' to', 'timestamp': (1.64, 1.84)},
                    {'text': ' its', 'timestamp': (1.84, 2.02)},
                    {'text': ' place', 'timestamp': (2.02, 2.28)},
                    {'text': ' amidst', 'timestamp': (2.28, 2.8)},
                    {'text': ' the', 'timestamp': (2.8, 2.98)},
                    {'text': ' tents.', 'timestamp': (2.98, 3.48)},
                ],
            },
        )
        # fmt: on

    @slow
    @require_torch
    def test_return_timestamps_and_language_in_preprocess(self):
        pipe = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-tiny",
            chunk_length_s=8,
            stride_length_s=1,
            return_language=True,
        )
        data = load_dataset("openslr/librispeech_asr", "clean", split="test", streaming=True, trust_remote_code=True)
        sample = next(iter(data))

        res = pipe(sample["audio"]["array"])
        self.assertEqual(
            res,
            {
                "text": " Conquered returned to its place amidst the tents.",
                "chunks": [{"language": "english", "text": " Conquered returned to its place amidst the tents."}],
            },
        )

        res = pipe(sample["audio"]["array"], return_timestamps=True)
        self.assertEqual(
            res,
            {
                "text": " Conquered returned to its place amidst the tents.",
                "chunks": [
                    {
                        "timestamp": (0.0, 3.36),
                        "language": "english",
                        "text": " Conquered returned to its place amidst the tents.",
                    }
                ],
            },
        )

        res = pipe(sample["audio"]["array"], return_timestamps="word")
        # fmt: off
        self.assertEqual(
            res,
            {
                'text': ' Conquered returned to its place amidst the tents.',
                'chunks': [
                    {"language": "english",'text': ' Conquered', 'timestamp': (0.5, 1.2)},
                    {"language": "english", 'text': ' returned', 'timestamp': (1.2, 1.64)},
                    {"language": "english",'text': ' to', 'timestamp': (1.64, 1.84)},
                    {"language": "english",'text': ' its', 'timestamp': (1.84, 2.02)},
                    {"language": "english",'text': ' place', 'timestamp': (2.02, 2.28)},
                    {"language": "english",'text': ' amidst', 'timestamp': (2.28, 2.8)},
                    {"language": "english",'text': ' the', 'timestamp': (2.8, 2.98)},
                    {"language": "english",'text': ' tents.', 'timestamp': (2.98, 3.48)},
                ],
            },
        )
        # fmt: on

    @slow
    @require_torch
    def test_return_timestamps_in_preprocess_longform(self):
        pipe = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-tiny.en",
        )
        data = load_dataset("openslr/librispeech_asr", "clean", split="test", streaming=True, trust_remote_code=True)
        samples = [next(iter(data)) for _ in range(8)]
        audio = np.concatenate([sample["audio"]["array"] for sample in samples])

        res = pipe(audio)
        expected_output = {
            "text": " Concord returned to its place amidst the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst "
            "the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst "
            "the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst "
            "the tents. Concord returned to its place amidst the tents."
        }
        self.assertEqual(res, expected_output)
        res = pipe(audio, return_timestamps=True)
        self.assertEqual(
            res,
            {
                "text": " Concord returned to its place amidst the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst the tents. Concord returned to its place amidst the tents.",
                "chunks": [
                    {"timestamp": (0.0, 3.22), "text": " Concord returned to its place amidst the tents."},
                    {"timestamp": (3.22, 6.74), "text": " Concord returned to its place amidst the tents."},
                    {"timestamp": (6.74, 10.26), "text": " Concord returned to its place amidst the tents."},
                    {"timestamp": (10.26, 13.78), "text": " Concord returned to its place amidst the tents."},
                    {"timestamp": (13.78, 17.3), "text": " Concord returned to its place amidst the tents."},
                    {"timestamp": (17.3, 20.82), "text": " Concord returned to its place amidst the tents."},
                    {"timestamp": (20.82, 24.34), "text": " Concord returned to its place amidst the tents."},
                    {"timestamp": (24.34, 27.86), "text": " Concord returned to its place amidst the tents."},
                ],
            },
        )
        pipe.model.generation_config.alignment_heads = [[2, 2], [3, 0], [3, 2], [3, 3], [3, 4], [3, 5]]
        res = pipe(audio, return_timestamps="word")

        # fmt: off
        self.assertEqual(
            res["chunks"][:15],
            [
                {"text": " Concord", "timestamp": (0.5, 0.94)},
                {"text": " returned", "timestamp": (0.94, 1.52)},
                {"text": " to", "timestamp": (1.52, 1.78)},
                {"text": " its", "timestamp": (1.78, 1.98)},
                {"text": " place", "timestamp": (1.98, 2.16)},
                {"text": " amidst", "timestamp": (2.16, 2.5)},
                {"text": " the", "timestamp": (2.5, 2.9)},
                {"text": " tents.", "timestamp": (2.9, 4.2)},
                {"text": " Concord", "timestamp": (4.2, 4.5)},
                {"text": " returned", "timestamp": (4.5, 5.0)},
                {"text": " to", "timestamp": (5.0, 5.28)},
                {"text": " its", "timestamp": (5.28, 5.48)},
                {"text": " place", "timestamp": (5.48, 5.7)},
                {"text": " amidst", "timestamp": (5.7, 6.02)},
                {"text": " the", "timestamp": (6.02, 6.4)}


            ],
        )
        # fmt: on

    @require_torch
    def test_return_timestamps_in_init(self):
        # segment-level timestamps are accepted
        model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny")
        tokenizer = AutoTokenizer.from_pretrained("openai/whisper-tiny")
        feature_extractor = AutoFeatureExtractor.from_pretrained("openai/whisper-tiny")

        dummy_speech = np.ones(100)

        pipe = pipeline(
            task="automatic-speech-recognition",
            model=model,
            feature_extractor=feature_extractor,
            tokenizer=tokenizer,
            chunk_length_s=8,
            stride_length_s=1,
            return_timestamps=True,
        )

        _ = pipe(dummy_speech)

        # word-level timestamps are accepted
        pipe = pipeline(
            task="automatic-speech-recognition",
            model=model,
            feature_extractor=feature_extractor,
            tokenizer=tokenizer,
            chunk_length_s=8,
            stride_length_s=1,
            return_timestamps="word",
        )

        _ = pipe(dummy_speech)

        # char-level timestamps are not accepted
        with self.assertRaisesRegex(
            ValueError,
            "^Whisper cannot return `char` timestamps, only word level or segment level timestamps. "
            "Use `return_timestamps='word'` or `return_timestamps=True` respectively.$",
        ):
            pipe = pipeline(
                task="automatic-speech-recognition",
                model=model,
                feature_extractor=feature_extractor,
                tokenizer=tokenizer,
                chunk_length_s=8,
                stride_length_s=1,
                return_timestamps="char",
            )

            _ = pipe(dummy_speech)

    @require_torch
    @slow
    def test_torch_whisper(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-tiny",
            framework="pt",
        )
        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = speech_recognizer(audio)
        self.assertEqual(output, {"text": " A man said to the universe, Sir, I exist."})

        output = speech_recognizer([ds[40]["audio"]], chunk_length_s=5, batch_size=4)
        self.assertEqual(output, [{"text": " A man said to the universe, Sir, I exist."}])

    @require_torch
    @slow
    def test_torch_whisper_batched(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-tiny",
            framework="pt",
        )
        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation[:2]")
        EXPECTED_OUTPUT = [
            {"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel."},
            {"text": " Nor is Mr. Quilters' manner less interesting than his matter."},
        ]

        output = speech_recognizer(ds["audio"], batch_size=2)
        self.assertEqual(output, EXPECTED_OUTPUT)

    @slow
    def test_find_longest_common_subsequence(self):
        max_source_positions = 1500
        processor = AutoProcessor.from_pretrained("openai/whisper-tiny")

        previous_sequence = [[51492, 406, 3163, 1953, 466, 13, 51612, 51612]]
        self.assertEqual(
            processor.decode(previous_sequence[0], output_offsets=True),
            {
                "text": " not worth thinking about.",
                "offsets": [{"text": " not worth thinking about.", "timestamp": (22.56, 24.96)}],
            },
        )

        # Merge when the previous sequence is a suffix of the next sequence
        # fmt: off
        next_sequences_1 = [
            [50364, 295, 6177, 3391, 11, 19817, 3337, 507, 307, 406, 3163, 1953, 466, 13, 50614, 50614, 2812, 9836, 14783, 390, 6263, 538, 257, 1359, 11, 8199, 6327, 1090, 322, 702, 7443, 13, 50834, 50257]
        ]
        # fmt: on
        self.assertEqual(
            processor.decode(next_sequences_1[0], output_offsets=True),
            {
                "text": (
                    " of spectators, retrievality is not worth thinking about. His instant panic was followed by a"
                    " small, sharp blow high on his chest.<|endoftext|>"
                ),
                "offsets": [
                    {"text": " of spectators, retrievality is not worth thinking about.", "timestamp": (0.0, 5.0)},
                    {
                        "text": " His instant panic was followed by a small, sharp blow high on his chest.",
                        "timestamp": (5.0, 9.4),
                    },
                ],
            },
        )
        merge = _find_timestamp_sequence(
            [[previous_sequence, (480_000, 0, 0)], [next_sequences_1, (480_000, 120_000, 0)]],
            processor.tokenizer,
            processor.feature_extractor,
            max_source_positions,
        )

        # fmt: off
        self.assertEqual(
            merge,
            [51492, 406, 3163, 1953, 466, 13, 51739, 51739, 2812, 9836, 14783, 390, 6263, 538, 257, 1359, 11, 8199, 6327, 1090, 322, 702, 7443, 13, 51959],
        )
        # fmt: on
        self.assertEqual(
            processor.decode(merge, output_offsets=True),
            {
                "text": (
                    " not worth thinking about. His instant panic was followed by a small, sharp blow high on his"
                    " chest."
                ),
                "offsets": [
                    {"text": " not worth thinking about.", "timestamp": (22.56, 27.5)},
                    {
                        "text": " His instant panic was followed by a small, sharp blow high on his chest.",
                        "timestamp": (27.5, 31.900000000000002),
                    },
                ],
            },
        )

        # Merge when the sequence is in the middle of the 1st next sequence
        # fmt: off
        next_sequences_2 = [
            [50364, 295, 6177, 3391, 11, 19817, 3337, 507, 307, 406, 3163, 1953, 466, 13, 2812, 9836, 14783, 390, 6263, 538, 257, 1359, 11, 8199, 6327, 1090, 322, 702, 7443, 13, 50834, 50257]
        ]
        # fmt: on
        # {'text': ' of spectators, retrievality is not worth thinking about. His instant panic was followed by a small, sharp blow high on his chest.','timestamp': (0.0, 9.4)}
        merge = _find_timestamp_sequence(
            [[previous_sequence, (480_000, 0, 0)], [next_sequences_2, (480_000, 120_000, 0)]],
            processor.tokenizer,
            processor.feature_extractor,
            max_source_positions,
        )
        # fmt: off
        self.assertEqual(
            merge,
            [51492, 406, 3163, 1953, 466, 13, 2812, 9836, 14783, 390, 6263, 538, 257, 1359, 11, 8199, 6327, 1090, 322, 702, 7443, 13, 51959],
        )
        # fmt: on
        self.assertEqual(
            processor.decode(merge, output_offsets=True),
            {
                "text": (
                    " not worth thinking about. His instant panic was followed by a small, sharp blow high on his"
                    " chest."
                ),
                "offsets": [
                    {
                        "text": (
                            " not worth thinking about. His instant panic was followed by a small, sharp blow high on"
                            " his chest."
                        ),
                        "timestamp": (22.56, 31.900000000000002),
                    },
                ],
            },
        )

        # Merge when the previous sequence is not included in the current sequence
        next_sequences_3 = [[50364, 2812, 9836, 14783, 390, 6263, 538, 257, 1359, 11, 8199, 6327, 1090, 322, 702, 7443, 13, 50584, 50257]]  # fmt: skip
        # {'text': ' His instant panic was followed by a small, sharp blow high on his chest.','timestamp': (0.0, 9.4)}
        merge = _find_timestamp_sequence(
            [[previous_sequence, (480_000, 0, 0)], [next_sequences_3, (480_000, 120_000, 0)]],
            processor.tokenizer,
            processor.feature_extractor,
            max_source_positions,
        )
        self.assertEqual(
            merge,
            [51492, 406, 3163, 1953, 466, 13, 51612, 51612, 2812, 9836, 14783, 390, 6263, 538, 257, 1359, 11, 8199, 6327, 1090, 322, 702, 7443, 13, 51832],
        )  # fmt: skip
        self.assertEqual(
            processor.decode(merge, output_offsets=True),
            {
                "text": (
                    " not worth thinking about. His instant panic was followed by a small, sharp blow high on his"
                    " chest."
                ),
                "offsets": [
                    {"text": " not worth thinking about.", "timestamp": (22.56, 24.96)},
                    {
                        "text": " His instant panic was followed by a small, sharp blow high on his chest.",
                        "timestamp": (24.96, 29.36),
                    },
                ],
            },
        )
        # last case is when the sequence is not in the first next predicted start and end of timestamp
        next_sequences_3 = [
            [50364, 2812, 9836, 14783, 390, 406, 3163, 1953, 466, 13, 50634, 50634, 2812, 9836, 14783, 390, 6263, 538, 257, 1359, 11, 8199, 6327, 1090, 322, 702, 7443, 13, 50934]
        ]  # fmt: skip
        merge = _find_timestamp_sequence(
            [[previous_sequence, (480_000, 0, 0)], [next_sequences_3, (480_000, 167_000, 0)]],
            processor.tokenizer,
            processor.feature_extractor,
            max_source_positions,
        )
        self.assertEqual(
            merge,
            [51492, 406, 3163, 1953, 466, 13, 51612, 51612, 2812, 9836, 14783, 390, 6263, 538, 257, 1359, 11, 8199, 6327, 1090, 322, 702, 7443, 13, 51912]
        )  # fmt: skip
        self.assertEqual(
            processor.decode(merge, output_offsets=True),
            {
                "text": (
                    " not worth thinking about. His instant panic was followed by a small, sharp blow high on his"
                    " chest."
                ),
                "offsets": [
                    {"text": " not worth thinking about.", "timestamp": (22.56, 24.96)},
                    {
                        "text": " His instant panic was followed by a small, sharp blow high on his chest.",
                        "timestamp": (24.96, 30.96),
                    },
                ],
            },
        )

    @slow
    @require_torch
    def test_whisper_timestamp_prediction(self):
        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        array = np.concatenate(
            [ds[40]["audio"]["array"], ds[41]["audio"]["array"], ds[42]["audio"]["array"], ds[43]["audio"]["array"]]
        )
        pipe = pipeline(
            model="openai/whisper-small",
            return_timestamps=True,
        )

        output = pipe(ds[40]["audio"])
        self.assertDictEqual(
            output,
            {
                "text": " A man said to the universe, Sir, I exist.",
                "chunks": [{"text": " A man said to the universe, Sir, I exist.", "timestamp": (0.0, 4.26)}],
            },
        )

        output = pipe(array, chunk_length_s=10)
        self.assertDictEqual(
            nested_simplify(output),
            {
                "chunks": [
                    {"text": " A man said to the universe, Sir, I exist.", "timestamp": (0.0, 5.5)},
                    {
                        "text": (
                            " Sweat covered Brion's body, trickling into the "
                            "tight-loan cloth that was the only garment he wore, the "
                            "cut"
                        ),
                        "timestamp": (5.5, 11.95),
                    },
                    {
                        "text": (
                            " on his chest still dripping blood, the ache of his "
                            "overstrained eyes, even the soaring arena around him "
                            "with"
                        ),
                        "timestamp": (11.95, 19.61),
                    },
                    {
                        "text": " the thousands of spectators, retrievality is not worth thinking about.",
                        "timestamp": (19.61, 25.0),
                    },
                    {
                        "text": " His instant panic was followed by a small, sharp blow high on his chest.",
                        "timestamp": (25.0, 29.4),
                    },
                ],
                "text": (
                    " A man said to the universe, Sir, I exist. Sweat covered Brion's "
                    "body, trickling into the tight-loan cloth that was the only garment "
                    "he wore, the cut on his chest still dripping blood, the ache of his "
                    "overstrained eyes, even the soaring arena around him with the "
                    "thousands of spectators, retrievality is not worth thinking about. "
                    "His instant panic was followed by a small, sharp blow high on his "
                    "chest."
                ),
            },
        )

        output = pipe(array)
        self.assertDictEqual(
            output,
            {
                "chunks": [
                    {"text": " A man said to the universe, Sir, I exist.", "timestamp": (0.0, 5.5)},
                    {
                        "text": (
                            " Sweat covered Brion's body, trickling into the "
                            "tight-loan cloth that was the only garment"
                        ),
                        "timestamp": (5.5, 10.18),
                    },
                    {"text": " he wore.", "timestamp": (10.18, 11.68)},
                    {"text": " The cut on his chest still dripping blood.", "timestamp": (11.68, 14.92)},
                    {"text": " The ache of his overstrained eyes.", "timestamp": (14.92, 17.6)},
                    {
                        "text": (
                            " Even the soaring arena around him with the thousands of spectators were trivialities"
                        ),
                        "timestamp": (17.6, 22.56),
                    },
                    {"text": " not worth thinking about.", "timestamp": (22.56, 24.96)},
                ],
                "text": (
                    " A man said to the universe, Sir, I exist. Sweat covered Brion's "
                    "body, trickling into the tight-loan cloth that was the only garment "
                    "he wore. The cut on his chest still dripping blood. The ache of his "
                    "overstrained eyes. Even the soaring arena around him with the "
                    "thousands of spectators were trivialities not worth thinking about."
                ),
            },
        )

    @slow
    @require_torch
    def test_whisper_large_timestamp_prediction(self):
        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        array = np.concatenate(
            [ds[40]["audio"]["array"], ds[41]["audio"]["array"], ds[42]["audio"]["array"], ds[43]["audio"]["array"]]
        )
        pipe = pipeline(model="openai/whisper-large-v3", return_timestamps=True)

        output = pipe(ds[40]["audio"])
        self.assertDictEqual(
            output,
            {
                "text": " A man said to the universe, Sir, I exist.",
                "chunks": [{"text": " A man said to the universe, Sir, I exist.", "timestamp": (0.0, 4.08)}],
            },
        )

        output = pipe(array, chunk_length_s=10)

        self.assertDictEqual(
            nested_simplify(output),
            {
                "chunks": [
                    {"timestamp": (0.0, 2.0), "text": (" A man said to the universe,")},
                    {"timestamp": (2.0, 4.1), "text": (" Sir, I exist.")},
                    {"timestamp": (5.14, 5.96), "text": (" Sweat covered")},
                    {"timestamp": (5.96, 8.02), "text": (" Breon's body, trickling into")},
                    {"timestamp": (8.02, 10.67), "text": (" the tight loincloth that was the only garment he wore,")},
                    {"timestamp": (10.67, 13.67), "text": (" the cut on his chest still dripping blood,")},
                    {"timestamp": (13.67, 17.61), "text": (" the ache of his overstrained eyes.")},
                    {
                        "timestamp": (17.61, 24.0),
                        "text": (
                            " Even the soaring arena around him with thousands of spectators were trivialities not worth thinking about."
                        ),
                    },
                    {
                        "timestamp": (24.0, 29.94),
                        "text": (" His instant of panic was followed by a small, sharp blow high on his chest."),
                    },
                ],
                "text": (
                    " A man said to the universe, Sir, I exist. Sweat covered Breon's"
                    " body, trickling into the tight loincloth that was the only garment"
                    " he wore, the cut on his chest still dripping blood, the ache of his"
                    " overstrained eyes. Even the soaring arena around him with thousands"
                    " of spectators were trivialities not worth thinking about. His "
                    "instant of panic was followed by a small, sharp blow high on his chest."
                ),
            },
        )

        output = pipe(array)
        self.assertDictEqual(
            output,
            {
                "chunks": [
                    {"timestamp": (0.0, 1.96), "text": " A man said to the universe,"},
                    {"timestamp": (2.7, 4.1), "text": " Sir, I exist."},
                    {"timestamp": (5.14, 6.84), "text": " Sweat covered Brion's body,"},
                    {
                        "timestamp": (7.4, 10.68),
                        "text": " trickling into the tight loincloth that was the only garment he wore,",
                    },
                    {"timestamp": (11.6, 13.94), "text": " the cut on his chest still dripping blood,"},
                    {"timestamp": (14.78, 16.72), "text": " the ache of his overstrained eyes,"},
                    {
                        "timestamp": (17.32, 21.16),
                        "text": " even the soaring arena around him with the thousands of spectators",
                    },
                    {"timestamp": (21.16, 23.94), "text": " were trivialities not worth thinking about."},
                    {
                        "timestamp": (24.42, 29.94),
                        "text": " His instant panic was followed by a small sharp blow high on his chest.",
                    },
                ],
                "text": (
                    " A man said to the universe, Sir, I exist. Sweat covered Brion's body,"
                    " trickling into the tight loincloth that was the only garment he wore, "
                    "the cut on his chest still dripping blood, the ache of his overstrained "
                    "eyes, even the soaring arena around him with the thousands of spectators "
                    "were trivialities not worth thinking about. His instant panic was followed "
                    "by a small sharp blow high on his chest."
                ),
            },
        )

    @slow
    @require_torch
    def test_whisper_word_timestamps_batched(self):
        pipe = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-tiny",
            chunk_length_s=3,
            return_timestamps="word",
        )
        data = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
        sample = data[0]["audio"]

        # not the same output as test_simple_whisper_asr because of chunking
        EXPECTED_OUTPUT = {
            "text": " Mr. Quilder is the apostle of the middle classes and we are glad to welcome his gospel.",
            "chunks": [
                {"text": " Mr.", "timestamp": (0.48, 0.96)},
                {"text": " Quilder", "timestamp": (0.96, 1.24)},
                {"text": " is", "timestamp": (1.24, 1.5)},
                {"text": " the", "timestamp": (1.5, 1.72)},
                {"text": " apostle", "timestamp": (1.72, 1.98)},
                {"text": " of", "timestamp": (1.98, 2.32)},
                {"text": " the", "timestamp": (2.32, 2.5)},
                {"text": " middle", "timestamp": (2.5, 2.68)},
                {"text": " classes", "timestamp": (2.68, 3.2)},
                {"text": " and", "timestamp": (3.2, 3.56)},
                {"text": " we", "timestamp": (3.56, 3.68)},
                {"text": " are", "timestamp": (3.68, 3.8)},
                {"text": " glad", "timestamp": (3.8, 4.1)},
                {"text": " to", "timestamp": (4.1, 4.34)},
                {"text": " welcome", "timestamp": (4.3, 4.6)},
                {"text": " his", "timestamp": (4.6, 4.94)},
                {"text": " gospel.", "timestamp": (4.94, 5.82)},
            ],
        }

        # batch size 1: copy the audio sample since pipeline consumes it
        output = pipe(sample.copy(), batch_size=1)
        self.assertDictEqual(output, EXPECTED_OUTPUT)

        # batch size 2: input audio is chunked into smaller pieces so it's testing batching
        output = pipe(sample, batch_size=2)
        self.assertDictEqual(output, EXPECTED_OUTPUT)

    @slow
    @require_torch
    def test_whisper_large_word_timestamps_batched(self):
        pipe = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-large-v3",
            return_timestamps="word",
        )
        data = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
        sample = data[0]["audio"]

        # not the same output as test_simple_whisper_asr because of chunking
        EXPECTED_OUTPUT = {
            "text": " Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.",
            "chunks": [
                {"text": " Mr.", "timestamp": (0.0, 0.74)},
                {"text": " Quilter", "timestamp": (0.74, 1.04)},
                {"text": " is", "timestamp": (1.04, 1.3)},
                {"text": " the", "timestamp": (1.3, 1.44)},
                {"text": " apostle", "timestamp": (1.44, 1.74)},
                {"text": " of", "timestamp": (1.74, 2.18)},
                {"text": " the", "timestamp": (2.18, 2.28)},
                {"text": " middle", "timestamp": (2.28, 2.5)},
                {"text": " classes,", "timestamp": (2.5, 3.0)},
                {"text": " and", "timestamp": (3.0, 3.4)},
                {"text": " we", "timestamp": (3.4, 3.5)},
                {"text": " are", "timestamp": (3.5, 3.6)},
                {"text": " glad", "timestamp": (3.6, 3.84)},
                {"text": " to", "timestamp": (3.84, 4.1)},
                {"text": " welcome", "timestamp": (4.1, 4.4)},
                {"text": " his", "timestamp": (4.4, 4.7)},
                {"text": " gospel.", "timestamp": (4.7, 5.34)},
            ],
        }

        # batch size 1: copy the audio sample since pipeline consumes it
        output = pipe(sample.copy(), batch_size=1)
        self.assertDictEqual(output, EXPECTED_OUTPUT)

        # batch size 2: input audio is chunked into smaller pieces so it's testing batching
        output = pipe(sample, batch_size=2)
        self.assertDictEqual(output, EXPECTED_OUTPUT)

    @require_torch
    @slow
    def test_torch_speech_encoder_decoder(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="facebook/s2t-wav2vec2-large-en-de",
            feature_extractor="facebook/s2t-wav2vec2-large-en-de",
            framework="pt",
        )

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = speech_recognizer(audio)
        self.assertEqual(output, {"text": 'Ein Mann sagte zum Universum : " Sir, ich existiert! "'})

    @slow
    @require_torch
    def test_simple_wav2vec2(self):
        model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")
        tokenizer = AutoTokenizer.from_pretrained("facebook/wav2vec2-base-960h")
        feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-base-960h")

        asr = AutomaticSpeechRecognitionPipeline(model=model, tokenizer=tokenizer, feature_extractor=feature_extractor)

        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)
        output = asr(waveform)
        self.assertEqual(output, {"text": ""})

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = asr(audio)
        self.assertEqual(output, {"text": "A MAN SAID TO THE UNIVERSE SIR I EXIST"})

        data = Audio().encode_example(ds[40]["audio"])["bytes"]
        output = asr(data)
        self.assertEqual(output, {"text": "A MAN SAID TO THE UNIVERSE SIR I EXIST"})

    @slow
    @require_torch
    @require_torchaudio
    def test_simple_s2t(self):
        model = Speech2TextForConditionalGeneration.from_pretrained("facebook/s2t-small-mustc-en-it-st")
        tokenizer = AutoTokenizer.from_pretrained("facebook/s2t-small-mustc-en-it-st")
        feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/s2t-small-mustc-en-it-st")

        asr = AutomaticSpeechRecognitionPipeline(model=model, tokenizer=tokenizer, feature_extractor=feature_extractor)

        waveform = np.tile(np.arange(1000, dtype=np.float32), 34)

        output = asr(waveform)
        self.assertEqual(output, {"text": "(Applausi)"})

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = asr(audio)
        self.assertEqual(output, {"text": "Un uomo disse all'universo: \"Signore, io esisto."})

        data = Audio().encode_example(ds[40]["audio"])["bytes"]
        output = asr(data)
        self.assertEqual(output, {"text": "Un uomo disse all'universo: \"Signore, io esisto."})

    @slow
    @require_torch
    @require_torchaudio
    def test_simple_whisper_asr(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-tiny.en",
            framework="pt",
        )
        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
        audio = ds[0]["audio"]
        output = speech_recognizer(audio)
        self.assertEqual(
            output,
            {"text": " Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel."},
        )
        output = speech_recognizer(ds[0]["audio"], return_timestamps=True)
        self.assertEqual(
            output,
            {
                "text": " Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.",
                "chunks": [
                    {
                        "text": (
                            " Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel."
                        ),
                        "timestamp": (0.0, 5.44),
                    }
                ],
            },
        )
        speech_recognizer.model.generation_config.alignment_heads = [[2, 2], [3, 0], [3, 2], [3, 3], [3, 4], [3, 5]]
        output = speech_recognizer(ds[0]["audio"], return_timestamps="word")
        # fmt: off
        self.assertEqual(
            output,
            {
                'text': ' Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.',
                'chunks': [
                    {'text': ' Mr.', 'timestamp': (0.38, 1.04)},
                    {'text': ' Quilter', 'timestamp': (1.04, 1.18)},
                    {'text': ' is', 'timestamp': (1.18, 1.44)},
                    {'text': ' the', 'timestamp': (1.44, 1.58)},
                    {'text': ' apostle', 'timestamp': (1.58, 1.98)},
                    {'text': ' of', 'timestamp': (1.98, 2.32)},
                    {'text': ' the', 'timestamp': (2.32, 2.46)},
                    {'text': ' middle', 'timestamp': (2.46, 2.56)},
                    {'text': ' classes,', 'timestamp': (2.56, 3.4)},
                    {'text': ' and', 'timestamp': (3.4, 3.54)},
                    {'text': ' we', 'timestamp': (3.54, 3.62)},
                    {'text': ' are', 'timestamp': (3.62, 3.72)},
                    {'text': ' glad', 'timestamp': (3.72, 4.0)},
                    {'text': ' to', 'timestamp': (4.0, 4.26)},
                    {'text': ' welcome', 'timestamp': (4.26, 4.56)},
                    {'text': ' his', 'timestamp': (4.56, 4.92)},
                    {'text': ' gospel.', 'timestamp': (4.92, 5.84)}
                ]
            }
        )
        # fmt: on

        # Whisper can only predict segment level timestamps or word level, not character level
        with self.assertRaisesRegex(
            ValueError,
            "^Whisper cannot return `char` timestamps, only word level or segment level timestamps. "
            "Use `return_timestamps='word'` or `return_timestamps=True` respectively.$",
        ):
            _ = speech_recognizer(audio, return_timestamps="char")

    @slow
    @require_torch
    @require_torchaudio
    def test_simple_whisper_translation(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-large",
            framework="pt",
        )
        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = speech_recognizer(audio)
        self.assertEqual(output, {"text": " A man said to the universe, Sir, I exist."})

        model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large")
        tokenizer = AutoTokenizer.from_pretrained("openai/whisper-large")
        feature_extractor = AutoFeatureExtractor.from_pretrained("openai/whisper-large")

        speech_recognizer_2 = AutomaticSpeechRecognitionPipeline(
            model=model, tokenizer=tokenizer, feature_extractor=feature_extractor
        )
        output_2 = speech_recognizer_2(ds[40]["audio"])
        self.assertEqual(output, output_2)

        # either use generate_kwargs or set the model's generation_config
        # model.generation_config.task = "transcribe"
        # model.generation_config.lang = "<|it|>"
        speech_translator = AutomaticSpeechRecognitionPipeline(
            model=model,
            tokenizer=tokenizer,
            feature_extractor=feature_extractor,
            generate_kwargs={"task": "transcribe", "language": "<|it|>"},
        )
        output_3 = speech_translator(ds[40]["audio"])
        self.assertEqual(output_3, {"text": " Un uomo ha detto all'universo, Sir, esiste."})

    @slow
    @require_torch
    def test_whisper_language(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-tiny.en",
            framework="pt",
        )
        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
        audio = ds[0]["audio"]

        # 1. English-only model compatible with no language argument
        output = speech_recognizer(audio)
        self.assertEqual(
            output,
            {"text": " Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel."},
        )

        # 2. English-only Whisper does not accept the language argument
        with self.assertRaisesRegex(
            ValueError,
            "Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, "
            "pass `is_multilingual=True` to generate, or update the generation config.",
        ):
            _ = speech_recognizer(ds[0]["audio"], generate_kwargs={"language": "en"})

        # 3. Multilingual model accepts language argument
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="openai/whisper-tiny",
            framework="pt",
        )
        output = speech_recognizer(ds[0]["audio"], generate_kwargs={"language": "en"})
        self.assertEqual(
            output,
            {"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel."},
        )

    @slow
    def test_speculative_decoding_whisper_non_distil(self):
        # Load data:
        dataset = load_dataset(
            "hf-internal-testing/librispeech_asr_dummy", "clean", split="validation[:1]", trust_remote_code=True
        )
        sample = dataset[0]["audio"]

        # Load model:
        model_id = "openai/whisper-large-v2"
        processor = AutoProcessor.from_pretrained(model_id)
        model = AutoModelForSpeechSeq2Seq.from_pretrained(
            model_id,
            use_safetensors=True,
        )

        # Load assistant:
        assistant_model_id = "openai/whisper-tiny"
        assistant_model = AutoModelForSpeechSeq2Seq.from_pretrained(
            assistant_model_id,
            use_safetensors=True,
        )

        # Load pipeline:
        pipe = AutomaticSpeechRecognitionPipeline(
            model=model,
            tokenizer=processor.tokenizer,
            feature_extractor=processor.feature_extractor,
            generate_kwargs={"language": "en"},
        )

        start_time = time.time()
        transcription_non_ass = pipe(sample.copy(), generate_kwargs={"assistant_model": assistant_model})["text"]
        total_time_assist = time.time() - start_time

        start_time = time.time()
        transcription_ass = pipe(sample)["text"]
        total_time_non_assist = time.time() - start_time

        self.assertEqual(transcription_ass, transcription_non_ass)
        self.assertEqual(
            transcription_ass,
            " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
        )
        self.assertTrue(total_time_non_assist > total_time_assist, "Make sure that assistant decoding is faster")

    @slow
    def test_speculative_decoding_whisper_distil(self):
        # Load data:
        dataset = load_dataset(
            "hf-internal-testing/librispeech_asr_dummy", "clean", split="validation[:1]", trust_remote_code=True
        )
        sample = dataset[0]["audio"]

        # Load model:
        model_id = "openai/whisper-large-v2"
        processor = AutoProcessor.from_pretrained(model_id)
        model = AutoModelForSpeechSeq2Seq.from_pretrained(
            model_id,
            use_safetensors=True,
        )

        # Load assistant:
        assistant_model_id = "distil-whisper/distil-large-v2"
        assistant_model = AutoModelForCausalLM.from_pretrained(
            assistant_model_id,
            use_safetensors=True,
        )

        # Load pipeline:
        pipe = AutomaticSpeechRecognitionPipeline(
            model=model,
            tokenizer=processor.tokenizer,
            feature_extractor=processor.feature_extractor,
            generate_kwargs={"language": "en"},
        )

        start_time = time.time()
        transcription_non_ass = pipe(sample.copy(), generate_kwargs={"assistant_model": assistant_model})["text"]
        total_time_assist = time.time() - start_time

        start_time = time.time()
        transcription_ass = pipe(sample)["text"]
        total_time_non_assist = time.time() - start_time

        self.assertEqual(transcription_ass, transcription_non_ass)
        self.assertEqual(
            transcription_ass,
            " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
        )
        self.assertEqual(total_time_non_assist > total_time_assist, "Make sure that assistant decoding is faster")

    @slow
    @require_torch
    @require_torchaudio
    def test_xls_r_to_en(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="facebook/wav2vec2-xls-r-1b-21-to-en",
            feature_extractor="facebook/wav2vec2-xls-r-1b-21-to-en",
            framework="pt",
        )

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = speech_recognizer(audio)
        self.assertEqual(output, {"text": "A man said to the universe: “Sir, I exist."})

    @slow
    @require_torch
    @require_torchaudio
    def test_xls_r_from_en(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="facebook/wav2vec2-xls-r-1b-en-to-15",
            feature_extractor="facebook/wav2vec2-xls-r-1b-en-to-15",
            framework="pt",
        )

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = speech_recognizer(audio)
        self.assertEqual(output, {"text": "Ein Mann sagte zu dem Universum, Sir, ich bin da."})

    @slow
    @require_torch
    @require_torchaudio
    def test_speech_to_text_leveraged(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="patrickvonplaten/wav2vec2-2-bart-base",
            feature_extractor="patrickvonplaten/wav2vec2-2-bart-base",
            tokenizer=AutoTokenizer.from_pretrained("patrickvonplaten/wav2vec2-2-bart-base"),
            framework="pt",
        )

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]
        output = speech_recognizer(audio)
        self.assertEqual(output, {"text": "a man said to the universe sir i exist"})

    @slow
    @require_torch_accelerator
    def test_wav2vec2_conformer_float16(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="facebook/wav2vec2-conformer-rope-large-960h-ft",
            device=torch_device,
            torch_dtype=torch.float16,
            framework="pt",
        )

        dataset = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
        sample = dataset[0]["audio"]

        output = speech_recognizer(sample)
        self.assertEqual(
            output,
            {"text": "MISTER QUILTER IS THE APOSTLE OF THE MIDDLE CLASSES AND WE ARE GLAD TO WELCOME HIS GOSPEL"},
        )

    @require_torch
    def test_chunking_fast(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="hf-internal-testing/tiny-random-wav2vec2",
            chunk_length_s=10.0,
        )

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]["array"]

        n_repeats = 2
        audio_tiled = np.tile(audio, n_repeats)
        output = speech_recognizer([audio_tiled], batch_size=2)
        self.assertEqual(output, [{"text": ANY(str)}])
        self.assertEqual(output[0]["text"][:6], "ZBT ZC")

    @require_torch
    def test_input_parameter_passthrough(self):
        """Test that chunked vs non chunked versions of ASR pipelines returns the same structure for the same inputs."""
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="hf-internal-testing/tiny-random-wav2vec2",
        )

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]["array"]

        inputs = {"raw": audio, "sampling_rate": 16_000, "id": 1}

        chunked_output = speech_recognizer(inputs.copy(), chunk_length_s=30)
        non_chunked_output = speech_recognizer(inputs.copy())
        assert chunked_output.keys() == non_chunked_output.keys(), (
            "The output structure should be the same for chunked vs non-chunked versions of asr pipelines."
        )

    @require_torch
    def test_return_timestamps_ctc_fast(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="hf-internal-testing/tiny-random-wav2vec2",
        )

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        # Take short audio to keep the test readable
        audio = ds[40]["audio"]["array"][:800]

        output = speech_recognizer(audio, return_timestamps="char")
        self.assertEqual(
            output,
            {
                "text": "ZBT ZX G",
                "chunks": [
                    {"text": " ", "timestamp": (0.0, 0.012)},
                    {"text": "Z", "timestamp": (0.012, 0.016)},
                    {"text": "B", "timestamp": (0.016, 0.02)},
                    {"text": "T", "timestamp": (0.02, 0.024)},
                    {"text": " ", "timestamp": (0.024, 0.028)},
                    {"text": "Z", "timestamp": (0.028, 0.032)},
                    {"text": "X", "timestamp": (0.032, 0.036)},
                    {"text": " ", "timestamp": (0.036, 0.04)},
                    {"text": "G", "timestamp": (0.04, 0.044)},
                ],
            },
        )

        output = speech_recognizer(audio, return_timestamps="word")
        self.assertEqual(
            output,
            {
                "text": "ZBT ZX G",
                "chunks": [
                    {"text": "ZBT", "timestamp": (0.012, 0.024)},
                    {"text": "ZX", "timestamp": (0.028, 0.036)},
                    {"text": "G", "timestamp": (0.04, 0.044)},
                ],
            },
        )

    @require_torch
    @require_pyctcdecode
    def test_chunking_fast_with_lm(self):
        speech_recognizer = pipeline(
            model="hf-internal-testing/processor_with_lm",
            chunk_length_s=10.0,
        )

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]["array"]

        n_repeats = 2
        audio_tiled = np.tile(audio, n_repeats)
        # Batch_size = 1
        output1 = speech_recognizer([audio_tiled], batch_size=1)
        self.assertEqual(output1, [{"text": ANY(str)}])
        self.assertEqual(output1[0]["text"][:6], "<s> <s")

        # batch_size = 2
        output2 = speech_recognizer([audio_tiled], batch_size=2)
        self.assertEqual(output2, [{"text": ANY(str)}])
        self.assertEqual(output2[0]["text"][:6], "<s> <s")

        # TODO There is an offby one error because of the ratio.
        # Maybe logits get affected by the padding on this random
        # model is more likely. Add some masking ?
        # self.assertEqual(output1, output2)

    @require_torch
    @require_pyctcdecode
    def test_with_lm_fast(self):
        speech_recognizer = pipeline(
            model="hf-internal-testing/processor_with_lm",
        )
        self.assertEqual(speech_recognizer.type, "ctc_with_lm")

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]["array"]

        n_repeats = 2
        audio_tiled = np.tile(audio, n_repeats)

        output = speech_recognizer([audio_tiled], batch_size=2)
        self.assertEqual(output, [{"text": ANY(str)}])
        self.assertEqual(output[0]["text"][:6], "<s> <s")

        # Making sure the argument are passed to the decoder
        # Since no change happens in the result, check the error comes from
        # the `decode_beams` function.
        with self.assertRaises(TypeError) as e:
            output = speech_recognizer([audio_tiled], decoder_kwargs={"num_beams": 2})
            self.assertContains(e.msg, "TypeError: decode_beams() got an unexpected keyword argument 'num_beams'")
        output = speech_recognizer([audio_tiled], decoder_kwargs={"beam_width": 2})

    @require_torch
    @require_pyctcdecode
    def test_with_local_lm_fast(self):
        local_dir = snapshot_download("hf-internal-testing/processor_with_lm")
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model=local_dir,
        )
        self.assertEqual(speech_recognizer.type, "ctc_with_lm")

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]["array"]

        n_repeats = 2
        audio_tiled = np.tile(audio, n_repeats)

        output = speech_recognizer([audio_tiled], batch_size=2)

        self.assertEqual(output, [{"text": ANY(str)}])
        self.assertEqual(output[0]["text"][:6], "<s> <s")

    @require_torch
    @slow
    def test_whisper_prompted(self):
        processor = AutoProcessor.from_pretrained("openai/whisper-tiny")
        model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny")
        model = model.to(torch_device)

        pipe = pipeline(
            "automatic-speech-recognition",
            model=model,
            tokenizer=processor.tokenizer,
            feature_extractor=processor.feature_extractor,
            max_new_tokens=128,
            chunk_length_s=30,
            batch_size=16,
        )

        dataset = load_dataset("distil-whisper/librispeech_long", "clean", split="validation")
        sample = dataset[0]["audio"]

        # prompt the model to misspell "Mr Quilter" as "Mr Quillter"
        whisper_prompt = "Mr. Quillter."
        prompt_ids = pipe.tokenizer.get_prompt_ids(whisper_prompt, return_tensors="pt").to(torch_device)

        unprompted_result = pipe(sample.copy())["text"]
        prompted_result = pipe(sample, generate_kwargs={"prompt_ids": prompt_ids})["text"]

        # fmt: off
        EXPECTED_UNPROMPTED_RESULT = " Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel. Nor is Mr. Quilter's manner less interesting than his matter. He tells us that at this festive season of the year, with Christmas and roast beef looming before us, similarly drawn from eating and its results occur most readily to the mind. He has grave doubts whether Sir Frederick Latins work is really Greek after all and can discover in it but little of rocky Ithaca. Lennils, pictures are a sort of upguards and atom paintings and Mason's exquisite itals are as national as a jingo poem. Mr. Birkut Foster's landscapes smile at one much in the same way that Mr. Carker used to flash his teeth. And Mr. John Collier gives his sitter a cheerful slap on the back before he says like a shampoo or a Turkish bath. Next man"
        EXPECTED_PROMPTED_RESULT = " Mr. Quillter is the apostle of the middle classes, and we are glad to welcome his gospel. Nor is Mr. Quillter's manner less interesting than his matter. He tells us that at this festive season of the year, with Christmas and roast beef looming before us, similarly drawn from eating and its results occur most readily to the mind. He has grave doubts whether Sir Frederick Latins work is really great after all, and can discover in it but little of rocky Ithaca. Lennils, pictures are a sort of upguards and atom paintings, and Mason's exquisite itals are as national as a jingo poem. Mr. Birkut Foster's landscapes smile at one much in the same way that Mr. Carker used to flash his teeth. Mr. John Collier gives his sitter a cheerful slap on the back before he says like a shampoo or a Turkish bath. Next man."
        # fmt: on

        self.assertEqual(unprompted_result, EXPECTED_UNPROMPTED_RESULT)
        self.assertEqual(prompted_result, EXPECTED_PROMPTED_RESULT)

    @require_torch
    @slow
    def test_whisper_longform(self):
        # fmt: off
        EXPECTED_RESULT = " Folks, if you watch the show, you know, I spent a lot of time right over there. Patiently and astutely scrutinizing the boxwood and mahogany chest set of the day's biggest stories developing the central headline pawns, definitely maneuvering an oso topical night to F6, fainting a classic Sicilian, nade door variation on the news, all the while seeing eight moves deep and patiently marshalling the latest press releases into a fisher's shows in Lip Nitsky attack that culminates in the elegant lethal slow-played, all-passant checkmate that is my nightly monologue. But sometimes, sometimes, folks, I. CHEERING AND APPLAUSE Sometimes I startle away, cubside down in the monkey bars of a condemned playground on a super fun site. Get all hept up on goofballs. Rummage that were discarded tag bag of defective toys. Yank out a fist bowl of disembodied doll limbs, toss them on Saturday, Rusty Cargo, container down by the Wharf, and challenge toothless drifters to the godless bughouse lets of tournament that is my segment. MUSIC Meanwhile!"
        # fmt: on

        processor = AutoProcessor.from_pretrained("openai/whisper-tiny.en")
        model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
        model = model.to(torch_device)

        pipe = pipeline(
            "automatic-speech-recognition",
            model=model,
            tokenizer=processor.tokenizer,
            feature_extractor=processor.feature_extractor,
            max_new_tokens=128,
            device=torch_device,
            return_timestamps=True,  # to allow longform generation
        )

        ds = load_dataset("distil-whisper/meanwhile", "default")["test"]
        ds = ds.cast_column("audio", Audio(sampling_rate=16000))
        audio = ds[:1]["audio"]

        result = pipe(audio)[0]["text"]

        assert result == EXPECTED_RESULT

    @require_torch
    @slow
    def test_seamless_v2(self):
        pipe = pipeline(
            "automatic-speech-recognition",
            model="facebook/seamless-m4t-v2-large",
            device=torch_device,
        )

        dataset = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
        sample = dataset[0]["audio"]

        result = pipe(sample, generate_kwargs={"tgt_lang": "eng"})
        EXPECTED_RESULT = "mister quilter is the apostle of the middle classes and we are glad to welcome his gospel"

        assert result["text"] == EXPECTED_RESULT

    @require_torch
    @slow
    def test_chunking_and_timestamps(self):
        model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")
        tokenizer = AutoTokenizer.from_pretrained("facebook/wav2vec2-base-960h")
        feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-base-960h")
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model=model,
            tokenizer=tokenizer,
            feature_extractor=feature_extractor,
            framework="pt",
            chunk_length_s=10.0,
        )

        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]["array"]

        n_repeats = 10
        audio_tiled = np.tile(audio, n_repeats)
        output = speech_recognizer([audio_tiled], batch_size=2)
        self.assertEqual(output, [{"text": ("A MAN SAID TO THE UNIVERSE SIR I EXIST " * n_repeats).strip()}])

        output = speech_recognizer(audio, return_timestamps="char")
        self.assertEqual(audio.shape, (74_400,))
        self.assertEqual(speech_recognizer.feature_extractor.sampling_rate, 16_000)
        # The audio is 74_400 / 16_000 = 4.65s long.
        self.assertEqual(
            output,
            {
                "text": "A MAN SAID TO THE UNIVERSE SIR I EXIST",
                "chunks": [
                    {"text": "A", "timestamp": (0.6, 0.62)},
                    {"text": " ", "timestamp": (0.62, 0.66)},
                    {"text": "M", "timestamp": (0.68, 0.7)},
                    {"text": "A", "timestamp": (0.78, 0.8)},
                    {"text": "N", "timestamp": (0.84, 0.86)},
                    {"text": " ", "timestamp": (0.92, 0.98)},
                    {"text": "S", "timestamp": (1.06, 1.08)},
                    {"text": "A", "timestamp": (1.14, 1.16)},
                    {"text": "I", "timestamp": (1.16, 1.18)},
                    {"text": "D", "timestamp": (1.2, 1.24)},
                    {"text": " ", "timestamp": (1.24, 1.28)},
                    {"text": "T", "timestamp": (1.28, 1.32)},
                    {"text": "O", "timestamp": (1.34, 1.36)},
                    {"text": " ", "timestamp": (1.38, 1.42)},
                    {"text": "T", "timestamp": (1.42, 1.44)},
                    {"text": "H", "timestamp": (1.44, 1.46)},
                    {"text": "E", "timestamp": (1.46, 1.5)},
                    {"text": " ", "timestamp": (1.5, 1.56)},
                    {"text": "U", "timestamp": (1.58, 1.62)},
                    {"text": "N", "timestamp": (1.64, 1.68)},
                    {"text": "I", "timestamp": (1.7, 1.72)},
                    {"text": "V", "timestamp": (1.76, 1.78)},
                    {"text": "E", "timestamp": (1.84, 1.86)},
                    {"text": "R", "timestamp": (1.86, 1.9)},
                    {"text": "S", "timestamp": (1.96, 1.98)},
                    {"text": "E", "timestamp": (1.98, 2.02)},
                    {"text": " ", "timestamp": (2.02, 2.06)},
                    {"text": "S", "timestamp": (2.82, 2.86)},
                    {"text": "I", "timestamp": (2.94, 2.96)},
                    {"text": "R", "timestamp": (2.98, 3.02)},
                    {"text": " ", "timestamp": (3.06, 3.12)},
                    {"text": "I", "timestamp": (3.5, 3.52)},
                    {"text": " ", "timestamp": (3.58, 3.6)},
                    {"text": "E", "timestamp": (3.66, 3.68)},
                    {"text": "X", "timestamp": (3.68, 3.7)},
                    {"text": "I", "timestamp": (3.9, 3.92)},
                    {"text": "S", "timestamp": (3.94, 3.96)},
                    {"text": "T", "timestamp": (4.0, 4.02)},
                    {"text": " ", "timestamp": (4.06, 4.1)},
                ],
            },
        )
        output = speech_recognizer(audio, return_timestamps="word")
        self.assertEqual(
            output,
            {
                "text": "A MAN SAID TO THE UNIVERSE SIR I EXIST",
                "chunks": [
                    {"text": "A", "timestamp": (0.6, 0.62)},
                    {"text": "MAN", "timestamp": (0.68, 0.86)},
                    {"text": "SAID", "timestamp": (1.06, 1.24)},
                    {"text": "TO", "timestamp": (1.28, 1.36)},
                    {"text": "THE", "timestamp": (1.42, 1.5)},
                    {"text": "UNIVERSE", "timestamp": (1.58, 2.02)},
                    {"text": "SIR", "timestamp": (2.82, 3.02)},
                    {"text": "I", "timestamp": (3.5, 3.52)},
                    {"text": "EXIST", "timestamp": (3.66, 4.02)},
                ],
            },
        )
        output = speech_recognizer(audio, return_timestamps="word", chunk_length_s=2.0)
        self.assertEqual(
            output,
            {
                "text": "A MAN SAID TO THE UNIVERSE SIR I EXIST",
                "chunks": [
                    {"text": "A", "timestamp": (0.6, 0.62)},
                    {"text": "MAN", "timestamp": (0.68, 0.86)},
                    {"text": "SAID", "timestamp": (1.06, 1.24)},
                    {"text": "TO", "timestamp": (1.3, 1.36)},
                    {"text": "THE", "timestamp": (1.42, 1.48)},
                    {"text": "UNIVERSE", "timestamp": (1.58, 2.02)},
                    # Tiny change linked to chunking.
                    {"text": "SIR", "timestamp": (2.84, 3.02)},
                    {"text": "I", "timestamp": (3.5, 3.52)},
                    {"text": "EXIST", "timestamp": (3.66, 4.02)},
                ],
            },
        )
        # CTC models must specify return_timestamps type - cannot set `return_timestamps=True` blindly
        with self.assertRaisesRegex(
            ValueError,
            "^CTC can either predict character level timestamps, or word level timestamps. "
            "Set `return_timestamps='char'` or `return_timestamps='word'` as required.$",
        ):
            _ = speech_recognizer(audio, return_timestamps=True)

    @require_torch
    @slow
    def test_chunking_with_lm(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="patrickvonplaten/wav2vec2-base-100h-with-lm",
            chunk_length_s=10.0,
        )
        ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation").sort("id")
        audio = ds[40]["audio"]["array"]

        n_repeats = 10
        audio = np.tile(audio, n_repeats)
        output = speech_recognizer([audio], batch_size=2)
        expected_text = "A MAN SAID TO THE UNIVERSE SIR I EXIST " * n_repeats
        expected = [{"text": expected_text.strip()}]
        self.assertEqual(output, expected)

    @require_torch
    def test_chunk_iterator(self):
        feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-base-960h")
        inputs = torch.arange(100).long()
        outs = list(chunk_iter(inputs, feature_extractor, 100, 0, 0))

        self.assertEqual(len(outs), 1)
        self.assertEqual([o["stride"] for o in outs], [(100, 0, 0)])
        self.assertEqual([o["input_values"].shape for o in outs], [(1, 100)])
        self.assertEqual([o["is_last"] for o in outs], [True])

        # two chunks no stride
        outs = list(chunk_iter(inputs, feature_extractor, 50, 0, 0))
        self.assertEqual(len(outs), 2)
        self.assertEqual([o["stride"] for o in outs], [(50, 0, 0), (50, 0, 0)])
        self.assertEqual([o["input_values"].shape for o in outs], [(1, 50), (1, 50)])
        self.assertEqual([o["is_last"] for o in outs], [False, True])

        # two chunks incomplete last
        outs = list(chunk_iter(inputs, feature_extractor, 80, 0, 0))
        self.assertEqual(len(outs), 2)
        self.assertEqual([o["stride"] for o in outs], [(80, 0, 0), (20, 0, 0)])
        self.assertEqual([o["input_values"].shape for o in outs], [(1, 80), (1, 20)])
        self.assertEqual([o["is_last"] for o in outs], [False, True])

        # one chunk since first is also last, because it contains only data
        # in the right strided part we just mark that part as non stride
        # This test is specifically crafted to trigger a bug if next chunk
        # would be ignored by the fact that all the data would be
        # contained in the strided left data.
        outs = list(chunk_iter(inputs, feature_extractor, 105, 5, 5))
        self.assertEqual(len(outs), 1)
        self.assertEqual([o["stride"] for o in outs], [(100, 0, 0)])
        self.assertEqual([o["input_values"].shape for o in outs], [(1, 100)])
        self.assertEqual([o["is_last"] for o in outs], [True])

    @require_torch
    def test_chunk_iterator_stride(self):
        feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-base-960h")
        inputs = torch.arange(100).long()
        input_values = feature_extractor(inputs, sampling_rate=feature_extractor.sampling_rate, return_tensors="pt")[
            "input_values"
        ]
        outs = list(chunk_iter(inputs, feature_extractor, 100, 20, 10))
        self.assertEqual(len(outs), 1)
        self.assertEqual([o["stride"] for o in outs], [(100, 0, 0)])
        self.assertEqual([o["input_values"].shape for o in outs], [(1, 100)])
        self.assertEqual([o["is_last"] for o in outs], [True])

        outs = list(chunk_iter(inputs, feature_extractor, 80, 20, 10))
        self.assertEqual(len(outs), 2)
        self.assertEqual([o["stride"] for o in outs], [(80, 0, 10), (50, 20, 0)])
        self.assertEqual([o["input_values"].shape for o in outs], [(1, 80), (1, 50)])
        self.assertEqual([o["is_last"] for o in outs], [False, True])

        outs = list(chunk_iter(inputs, feature_extractor, 90, 20, 0))
        self.assertEqual(len(outs), 2)
        self.assertEqual([o["stride"] for o in outs], [(90, 0, 0), (30, 20, 0)])
        self.assertEqual([o["input_values"].shape for o in outs], [(1, 90), (1, 30)])

        outs = list(chunk_iter(inputs, feature_extractor, 36, 6, 6))
        self.assertEqual(len(outs), 4)
        self.assertEqual([o["stride"] for o in outs], [(36, 0, 6), (36, 6, 6), (36, 6, 6), (28, 6, 0)])
        self.assertEqual([o["input_values"].shape for o in outs], [(1, 36), (1, 36), (1, 36), (1, 28)])

        inputs = torch.LongTensor([i % 2 for i in range(100)])
        input_values = feature_extractor(inputs, sampling_rate=feature_extractor.sampling_rate, return_tensors="pt")[
            "input_values"
        ]
        outs = list(chunk_iter(inputs, feature_extractor, 30, 5, 5))
        self.assertEqual(len(outs), 5)
        self.assertEqual([o["stride"] for o in outs], [(30, 0, 5), (30, 5, 5), (30, 5, 5), (30, 5, 5), (20, 5, 0)])
        self.assertEqual([o["input_values"].shape for o in outs], [(1, 30), (1, 30), (1, 30), (1, 30), (1, 20)])
        self.assertEqual([o["is_last"] for o in outs], [False, False, False, False, True])
        # (0, 25)
        self.assertEqual(nested_simplify(input_values[:, :30]), nested_simplify(outs[0]["input_values"]))
        # (25, 45)
        self.assertEqual(nested_simplify(input_values[:, 20:50]), nested_simplify(outs[1]["input_values"]))
        # (45, 65)
        self.assertEqual(nested_simplify(input_values[:, 40:70]), nested_simplify(outs[2]["input_values"]))
        # (65, 85)
        self.assertEqual(nested_simplify(input_values[:, 60:90]), nested_simplify(outs[3]["input_values"]))
        # (85, 100)
        self.assertEqual(nested_simplify(input_values[:, 80:100]), nested_simplify(outs[4]["input_values"]))

    @require_torch
    def test_stride(self):
        speech_recognizer = pipeline(
            task="automatic-speech-recognition",
            model="hf-internal-testing/tiny-random-wav2vec2",
        )
        waveform = np.tile(np.arange(1000, dtype=np.float32), 10)
        output = speech_recognizer({"raw": waveform, "stride": (0, 0), "sampling_rate": 16_000})
        self.assertEqual(output, {"text": "OB XB  B EB BB  B EB B OB X"})

        # 0 effective ids Just take the middle one
        output = speech_recognizer({"raw": waveform, "stride": (5000, 5000), "sampling_rate": 16_000})
        self.assertEqual(output, {"text": ""})

        # Only 1 arange.
        output = speech_recognizer({"raw": waveform, "stride": (0, 9000), "sampling_rate": 16_000})
        self.assertEqual(output, {"text": "OB"})

        # 2nd arange
        output = speech_recognizer({"raw": waveform, "stride": (1000, 8000), "sampling_rate": 16_000})
        self.assertEqual(output, {"text": "XB"})

    @slow
    @require_torch_accelerator
    def test_slow_unfinished_sequence(self):
        from transformers import GenerationConfig

        pipe = pipeline(
            "automatic-speech-recognition",
            model="vasista22/whisper-hindi-large-v2",
            device=torch_device,
        )

        # the audio is 4 seconds long
        audio = hf_hub_download("Narsil/asr_dummy", filename="hindi.ogg", repo_type="dataset")

        # Original model wasn't trained with timestamps and has incorrect generation config
        out = pipe(
            audio,
            return_timestamps=True,
            generate_kwargs={"generation_config": GenerationConfig.from_pretrained("openai/whisper-large-v2")},
        )
        self.assertEqual(
            out,
            {
                "text": "मिर्ची में कितने विभिन्न प्रजातियां हैं",
                "chunks": [{"timestamp": (0.58, None), "text": "मिर्ची में कितने विभिन्न प्रजातियां हैं"}],
            },
        )

    @require_torch
    def test_pipeline_assisted_generation(self):
        """Tests that we can run assisted generation in the pipeline"""
        model = "openai/whisper-tiny"
        pipe = pipeline("automatic-speech-recognition", model=model, assistant_model=model)

        # We can run the pipeline
        prompt = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation[:1]")["audio"]
        _ = pipe(prompt)

        # It is running assisted generation under the hood (e.g. flags incompatible with assisted gen will crash)
        with self.assertRaises(ValueError):
            _ = pipe(prompt, generate_kwargs={"num_beams": 2})


def require_ffmpeg(test_case):
    """
    Decorator marking a test that requires FFmpeg.

    These tests are skipped when FFmpeg isn't installed.

    """
    import subprocess

    try:
        subprocess.check_output(["ffmpeg", "-h"], stderr=subprocess.DEVNULL)
        return test_case
    except Exception:
        return unittest.skip(reason="test requires ffmpeg")(test_case)


def bytes_iter(chunk_size, chunks):
    for i in range(chunks):
        yield bytes(range(i * chunk_size, (i + 1) * chunk_size))


@require_ffmpeg
class AudioUtilsTest(unittest.TestCase):
    def test_chunk_bytes_iter_too_big(self):
        iter_ = iter(chunk_bytes_iter(bytes_iter(chunk_size=3, chunks=2), 10, stride=(0, 0)))
        self.assertEqual(next(iter_), {"raw": b"\x00\x01\x02\x03\x04\x05", "stride": (0, 0)})
        with self.assertRaises(StopIteration):
            next(iter_)

    def test_chunk_bytes_iter(self):
        iter_ = iter(chunk_bytes_iter(bytes_iter(chunk_size=3, chunks=2), 3, stride=(0, 0)))
        self.assertEqual(next(iter_), {"raw": b"\x00\x01\x02", "stride": (0, 0)})
        self.assertEqual(next(iter_), {"raw": b"\x03\x04\x05", "stride": (0, 0)})
        with self.assertRaises(StopIteration):
            next(iter_)

    def test_chunk_bytes_iter_stride(self):
        iter_ = iter(chunk_bytes_iter(bytes_iter(chunk_size=3, chunks=2), 3, stride=(1, 1)))
        self.assertEqual(next(iter_), {"raw": b"\x00\x01\x02", "stride": (0, 1)})
        self.assertEqual(next(iter_), {"raw": b"\x01\x02\x03", "stride": (1, 1)})
        self.assertEqual(next(iter_), {"raw": b"\x02\x03\x04", "stride": (1, 1)})
        # This is finished, but the chunk_bytes doesn't know it yet.
        self.assertEqual(next(iter_), {"raw": b"\x03\x04\x05", "stride": (1, 1)})
        self.assertEqual(next(iter_), {"raw": b"\x04\x05", "stride": (1, 0)})
        with self.assertRaises(StopIteration):
            next(iter_)

    def test_chunk_bytes_iter_stride_stream(self):
        iter_ = iter(chunk_bytes_iter(bytes_iter(chunk_size=3, chunks=2), 5, stride=(1, 1), stream=True))
        self.assertEqual(next(iter_), {"raw": b"\x00\x01\x02", "stride": (0, 0), "partial": True})
        self.assertEqual(next(iter_), {"raw": b"\x00\x01\x02\x03\x04", "stride": (0, 1), "partial": False})
        self.assertEqual(next(iter_), {"raw": b"\x03\x04\x05", "stride": (1, 0), "partial": False})
        with self.assertRaises(StopIteration):
            next(iter_)

        iter_ = iter(chunk_bytes_iter(bytes_iter(chunk_size=3, chunks=3), 5, stride=(1, 1), stream=True))
        self.assertEqual(next(iter_), {"raw": b"\x00\x01\x02", "stride": (0, 0), "partial": True})
        self.assertEqual(next(iter_), {"raw": b"\x00\x01\x02\x03\x04", "stride": (0, 1), "partial": False})
        self.assertEqual(next(iter_), {"raw": b"\x03\x04\x05\x06\x07", "stride": (1, 1), "partial": False})
        self.assertEqual(next(iter_), {"raw": b"\x06\x07\x08", "stride": (1, 0), "partial": False})
        with self.assertRaises(StopIteration):
            next(iter_)

        iter_ = iter(chunk_bytes_iter(bytes_iter(chunk_size=3, chunks=3), 10, stride=(1, 1), stream=True))
        self.assertEqual(next(iter_), {"raw": b"\x00\x01\x02", "stride": (0, 0), "partial": True})
        self.assertEqual(next(iter_), {"raw": b"\x00\x01\x02\x03\x04\x05", "stride": (0, 0), "partial": True})
        self.assertEqual(
            next(iter_), {"raw": b"\x00\x01\x02\x03\x04\x05\x06\x07\x08", "stride": (0, 0), "partial": True}
        )
        self.assertEqual(
            next(iter_), {"raw": b"\x00\x01\x02\x03\x04\x05\x06\x07\x08", "stride": (0, 0), "partial": False}
        )
        with self.assertRaises(StopIteration):
            next(iter_)

    def test_ffmpeg_no_additional_args(self):
        mic = ffmpeg_microphone_live(16000, 2.0)
        mic.close()

    def test_ffmpeg_additional_args(self):
        mic = ffmpeg_microphone_live(16000, 2.0, ffmpeg_additional_args=["-nostdin"])
        mic.close()