owensong commited on
Commit
ca09899
·
verified ·
1 Parent(s): 93a627c

Add disclosed utterance-level ASR robustness diagnostics

Browse files
Files changed (2) hide show
  1. README.md +3 -2
  2. evaluation/asr_robustness.json +928 -0
README.md CHANGED
@@ -67,8 +67,9 @@ The selected Inflect-Micro-v2 checkpoint scored **4.406 UTMOS22** and **1.23% se
67
  - All four systems synthesize the same 200 unseen English prompts.
68
  - Semantic WER uses Whisper-large-v3 and the disclosed English text normalizer.
69
  - UTMOS22 uses `tarepan/SpeechMOS` v1.2.0 with 5,000 paired bootstrap samples.
70
- - NISQA dimensions, signal diagnostics, categorized ASR failures, speaker consistency, and multi-seed robustness are under `evaluation/`.
71
- - Competitor results apply only to the named checkpoints and voices, not every configuration of those projects.
 
72
 
73
  </details>
74
 
 
67
  - All four systems synthesize the same 200 unseen English prompts.
68
  - Semantic WER uses Whisper-large-v3 and the disclosed English text normalizer.
69
  - UTMOS22 uses `tarepan/SpeechMOS` v1.2.0 with 5,000 paired bootstrap samples.
70
+ - NISQA dimensions, signal diagnostics, categorized ASR failures, speaker consistency, and multi-seed robustness are under `evaluation/`.
71
+ - `evaluation/asr_robustness.json` separates sentence failures, semantic character errors, tail WER, and category-level errors.
72
+ - Competitor results apply only to the named checkpoints and voices, not every configuration of those projects.
73
 
74
  </details>
75
 
evaluation/asr_robustness.json ADDED
@@ -0,0 +1,928 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "format": "inflect_tts_asr_robustness_summary_v1",
3
+ "source_format": "inflect_tts_multi_asr_wer_v1",
4
+ "asr": "whisper-large-v3",
5
+ "notes": [
6
+ "Character error rate is calculated over semantic-normalized ASR transcripts.",
7
+ "Sentence error rate counts any utterance with at least one semantic word edit.",
8
+ "These metrics measure ASR recoverability and do not replace human listening."
9
+ ],
10
+ "systems": {
11
+ "inflect-micro-v2-finalist": {
12
+ "overall": {
13
+ "utterances": 200,
14
+ "semantic_errors": 32,
15
+ "semantic_words": 2608,
16
+ "semantic_wer": 0.012269938650306749,
17
+ "failed_utterances": 21,
18
+ "sentence_error_rate": 0.105,
19
+ "exact_sentence_rate": 0.895,
20
+ "character_errors": 62,
21
+ "reference_characters": 13718,
22
+ "character_error_rate": 0.004519609272488701,
23
+ "mean_utterance_wer": 0.021512501387501385,
24
+ "p95_utterance_wer": 0.18181818181818182,
25
+ "worst_utterance_wer": 1.0
26
+ },
27
+ "categories": {
28
+ "conversational": {
29
+ "utterances": 25,
30
+ "semantic_errors": 0,
31
+ "semantic_words": 309,
32
+ "semantic_wer": 0.0,
33
+ "failed_utterances": 0,
34
+ "sentence_error_rate": 0.0,
35
+ "exact_sentence_rate": 1.0,
36
+ "character_errors": 0,
37
+ "reference_characters": 1501,
38
+ "character_error_rate": 0.0,
39
+ "mean_utterance_wer": 0.0,
40
+ "p95_utterance_wer": 0.0,
41
+ "worst_utterance_wer": 0.0
42
+ },
43
+ "descriptive": {
44
+ "utterances": 16,
45
+ "semantic_errors": 0,
46
+ "semantic_words": 226,
47
+ "semantic_wer": 0.0,
48
+ "failed_utterances": 0,
49
+ "sentence_error_rate": 0.0,
50
+ "exact_sentence_rate": 1.0,
51
+ "character_errors": 0,
52
+ "reference_characters": 1146,
53
+ "character_error_rate": 0.0,
54
+ "mean_utterance_wer": 0.0,
55
+ "p95_utterance_wer": 0.0,
56
+ "worst_utterance_wer": 0.0
57
+ },
58
+ "emotional": {
59
+ "utterances": 24,
60
+ "semantic_errors": 0,
61
+ "semantic_words": 329,
62
+ "semantic_wer": 0.0,
63
+ "failed_utterances": 0,
64
+ "sentence_error_rate": 0.0,
65
+ "exact_sentence_rate": 1.0,
66
+ "character_errors": 0,
67
+ "reference_characters": 1574,
68
+ "character_error_rate": 0.0,
69
+ "mean_utterance_wer": 0.0,
70
+ "p95_utterance_wer": 0.0,
71
+ "worst_utterance_wer": 0.0
72
+ },
73
+ "homographs": {
74
+ "utterances": 6,
75
+ "semantic_errors": 1,
76
+ "semantic_words": 70,
77
+ "semantic_wer": 0.014285714285714285,
78
+ "failed_utterances": 1,
79
+ "sentence_error_rate": 0.16666666666666666,
80
+ "exact_sentence_rate": 0.8333333333333334,
81
+ "character_errors": 1,
82
+ "reference_characters": 375,
83
+ "character_error_rate": 0.0026666666666666666,
84
+ "mean_utterance_wer": 0.016666666666666666,
85
+ "p95_utterance_wer": 0.07500000000000001,
86
+ "worst_utterance_wer": 0.1
87
+ },
88
+ "long": {
89
+ "utterances": 15,
90
+ "semantic_errors": 1,
91
+ "semantic_words": 432,
92
+ "semantic_wer": 0.0023148148148148147,
93
+ "failed_utterances": 1,
94
+ "sentence_error_rate": 0.06666666666666667,
95
+ "exact_sentence_rate": 0.9333333333333333,
96
+ "character_errors": 6,
97
+ "reference_characters": 2206,
98
+ "character_error_rate": 0.0027198549410698096,
99
+ "mean_utterance_wer": 0.0018518518518518517,
100
+ "p95_utterance_wer": 0.008333333333333304,
101
+ "worst_utterance_wer": 0.027777777777777776
102
+ },
103
+ "long_range": {
104
+ "utterances": 6,
105
+ "semantic_errors": 0,
106
+ "semantic_words": 121,
107
+ "semantic_wer": 0.0,
108
+ "failed_utterances": 0,
109
+ "sentence_error_rate": 0.0,
110
+ "exact_sentence_rate": 1.0,
111
+ "character_errors": 0,
112
+ "reference_characters": 744,
113
+ "character_error_rate": 0.0,
114
+ "mean_utterance_wer": 0.0,
115
+ "p95_utterance_wer": 0.0,
116
+ "worst_utterance_wer": 0.0
117
+ },
118
+ "names_places": {
119
+ "utterances": 6,
120
+ "semantic_errors": 4,
121
+ "semantic_words": 57,
122
+ "semantic_wer": 0.07017543859649122,
123
+ "failed_utterances": 3,
124
+ "sentence_error_rate": 0.5,
125
+ "exact_sentence_rate": 0.5,
126
+ "character_errors": 13,
127
+ "reference_characters": 397,
128
+ "character_error_rate": 0.0327455919395466,
129
+ "mean_utterance_wer": 0.07407407407407407,
130
+ "p95_utterance_wer": 0.19444444444444442,
131
+ "worst_utterance_wer": 0.2222222222222222
132
+ },
133
+ "narrative": {
134
+ "utterances": 24,
135
+ "semantic_errors": 1,
136
+ "semantic_words": 352,
137
+ "semantic_wer": 0.002840909090909091,
138
+ "failed_utterances": 1,
139
+ "sentence_error_rate": 0.041666666666666664,
140
+ "exact_sentence_rate": 0.9583333333333334,
141
+ "character_errors": 1,
142
+ "reference_characters": 1762,
143
+ "character_error_rate": 0.0005675368898978433,
144
+ "mean_utterance_wer": 0.002777777777777778,
145
+ "p95_utterance_wer": 0.0,
146
+ "worst_utterance_wer": 0.06666666666666667
147
+ },
148
+ "numbers": {
149
+ "utterances": 6,
150
+ "semantic_errors": 10,
151
+ "semantic_words": 64,
152
+ "semantic_wer": 0.15625,
153
+ "failed_utterances": 5,
154
+ "sentence_error_rate": 0.8333333333333334,
155
+ "exact_sentence_rate": 0.16666666666666663,
156
+ "character_errors": 18,
157
+ "reference_characters": 322,
158
+ "character_error_rate": 0.055900621118012424,
159
+ "mean_utterance_wer": 0.17462121212121215,
160
+ "p95_utterance_wer": 0.35625,
161
+ "worst_utterance_wer": 0.375
162
+ },
163
+ "punctuation": {
164
+ "utterances": 6,
165
+ "semantic_errors": 0,
166
+ "semantic_words": 65,
167
+ "semantic_wer": 0.0,
168
+ "failed_utterances": 0,
169
+ "sentence_error_rate": 0.0,
170
+ "exact_sentence_rate": 1.0,
171
+ "character_errors": 0,
172
+ "reference_characters": 402,
173
+ "character_error_rate": 0.0,
174
+ "mean_utterance_wer": 0.0,
175
+ "p95_utterance_wer": 0.0,
176
+ "worst_utterance_wer": 0.0
177
+ },
178
+ "question": {
179
+ "utterances": 15,
180
+ "semantic_errors": 0,
181
+ "semantic_words": 165,
182
+ "semantic_wer": 0.0,
183
+ "failed_utterances": 0,
184
+ "sentence_error_rate": 0.0,
185
+ "exact_sentence_rate": 1.0,
186
+ "character_errors": 0,
187
+ "reference_characters": 834,
188
+ "character_error_rate": 0.0,
189
+ "mean_utterance_wer": 0.0,
190
+ "p95_utterance_wer": 0.0,
191
+ "worst_utterance_wer": 0.0
192
+ },
193
+ "rare_words": {
194
+ "utterances": 6,
195
+ "semantic_errors": 3,
196
+ "semantic_words": 57,
197
+ "semantic_wer": 0.05263157894736842,
198
+ "failed_utterances": 2,
199
+ "sentence_error_rate": 0.3333333333333333,
200
+ "exact_sentence_rate": 0.6666666666666667,
201
+ "character_errors": 3,
202
+ "reference_characters": 500,
203
+ "character_error_rate": 0.006,
204
+ "mean_utterance_wer": 0.046969696969696974,
205
+ "p95_utterance_wer": 0.16136363636363635,
206
+ "worst_utterance_wer": 0.18181818181818182
207
+ },
208
+ "short": {
209
+ "utterances": 24,
210
+ "semantic_errors": 4,
211
+ "semantic_words": 105,
212
+ "semantic_wer": 0.0380952380952381,
213
+ "failed_utterances": 4,
214
+ "sentence_error_rate": 0.16666666666666666,
215
+ "exact_sentence_rate": 0.8333333333333334,
216
+ "character_errors": 10,
217
+ "reference_characters": 458,
218
+ "character_error_rate": 0.021834061135371178,
219
+ "mean_utterance_wer": 0.07083333333333333,
220
+ "p95_utterance_wer": 0.25,
221
+ "worst_utterance_wer": 1.0
222
+ },
223
+ "technical": {
224
+ "utterances": 21,
225
+ "semantic_errors": 8,
226
+ "semantic_words": 256,
227
+ "semantic_wer": 0.03125,
228
+ "failed_utterances": 4,
229
+ "sentence_error_rate": 0.19047619047619047,
230
+ "exact_sentence_rate": 0.8095238095238095,
231
+ "character_errors": 10,
232
+ "reference_characters": 1497,
233
+ "character_error_rate": 0.006680026720106881,
234
+ "mean_utterance_wer": 0.030193615907901622,
235
+ "p95_utterance_wer": 0.2,
236
+ "worst_utterance_wer": 0.2857142857142857
237
+ }
238
+ }
239
+ },
240
+ "inflect-nano-v2-finalist": {
241
+ "overall": {
242
+ "utterances": 200,
243
+ "semantic_errors": 40,
244
+ "semantic_words": 2608,
245
+ "semantic_wer": 0.015337423312883436,
246
+ "failed_utterances": 23,
247
+ "sentence_error_rate": 0.115,
248
+ "exact_sentence_rate": 0.885,
249
+ "character_errors": 102,
250
+ "reference_characters": 13718,
251
+ "character_error_rate": 0.007435486222481412,
252
+ "mean_utterance_wer": 0.021362234987234986,
253
+ "p95_utterance_wer": 0.18272727272727243,
254
+ "worst_utterance_wer": 0.4
255
+ },
256
+ "categories": {
257
+ "conversational": {
258
+ "utterances": 25,
259
+ "semantic_errors": 1,
260
+ "semantic_words": 309,
261
+ "semantic_wer": 0.003236245954692557,
262
+ "failed_utterances": 1,
263
+ "sentence_error_rate": 0.04,
264
+ "exact_sentence_rate": 0.96,
265
+ "character_errors": 5,
266
+ "reference_characters": 1501,
267
+ "character_error_rate": 0.0033311125916055963,
268
+ "mean_utterance_wer": 0.0036363636363636364,
269
+ "p95_utterance_wer": 0.0,
270
+ "worst_utterance_wer": 0.09090909090909091
271
+ },
272
+ "descriptive": {
273
+ "utterances": 16,
274
+ "semantic_errors": 0,
275
+ "semantic_words": 226,
276
+ "semantic_wer": 0.0,
277
+ "failed_utterances": 0,
278
+ "sentence_error_rate": 0.0,
279
+ "exact_sentence_rate": 1.0,
280
+ "character_errors": 0,
281
+ "reference_characters": 1146,
282
+ "character_error_rate": 0.0,
283
+ "mean_utterance_wer": 0.0,
284
+ "p95_utterance_wer": 0.0,
285
+ "worst_utterance_wer": 0.0
286
+ },
287
+ "emotional": {
288
+ "utterances": 24,
289
+ "semantic_errors": 0,
290
+ "semantic_words": 329,
291
+ "semantic_wer": 0.0,
292
+ "failed_utterances": 0,
293
+ "sentence_error_rate": 0.0,
294
+ "exact_sentence_rate": 1.0,
295
+ "character_errors": 0,
296
+ "reference_characters": 1574,
297
+ "character_error_rate": 0.0,
298
+ "mean_utterance_wer": 0.0,
299
+ "p95_utterance_wer": 0.0,
300
+ "worst_utterance_wer": 0.0
301
+ },
302
+ "homographs": {
303
+ "utterances": 6,
304
+ "semantic_errors": 4,
305
+ "semantic_words": 70,
306
+ "semantic_wer": 0.05714285714285714,
307
+ "failed_utterances": 2,
308
+ "sentence_error_rate": 0.3333333333333333,
309
+ "exact_sentence_rate": 0.6666666666666667,
310
+ "character_errors": 10,
311
+ "reference_characters": 375,
312
+ "character_error_rate": 0.02666666666666667,
313
+ "mean_utterance_wer": 0.05000000000000001,
314
+ "p95_utterance_wer": 0.17500000000000002,
315
+ "worst_utterance_wer": 0.2
316
+ },
317
+ "long": {
318
+ "utterances": 15,
319
+ "semantic_errors": 2,
320
+ "semantic_words": 432,
321
+ "semantic_wer": 0.004629629629629629,
322
+ "failed_utterances": 1,
323
+ "sentence_error_rate": 0.06666666666666667,
324
+ "exact_sentence_rate": 0.9333333333333333,
325
+ "character_errors": 5,
326
+ "reference_characters": 2206,
327
+ "character_error_rate": 0.0022665457842248413,
328
+ "mean_utterance_wer": 0.00404040404040404,
329
+ "p95_utterance_wer": 0.01818181818181812,
330
+ "worst_utterance_wer": 0.06060606060606061
331
+ },
332
+ "long_range": {
333
+ "utterances": 6,
334
+ "semantic_errors": 1,
335
+ "semantic_words": 121,
336
+ "semantic_wer": 0.008264462809917356,
337
+ "failed_utterances": 1,
338
+ "sentence_error_rate": 0.16666666666666666,
339
+ "exact_sentence_rate": 0.8333333333333334,
340
+ "character_errors": 2,
341
+ "reference_characters": 744,
342
+ "character_error_rate": 0.002688172043010753,
343
+ "mean_utterance_wer": 0.008333333333333333,
344
+ "p95_utterance_wer": 0.037500000000000006,
345
+ "worst_utterance_wer": 0.05
346
+ },
347
+ "names_places": {
348
+ "utterances": 6,
349
+ "semantic_errors": 5,
350
+ "semantic_words": 57,
351
+ "semantic_wer": 0.08771929824561403,
352
+ "failed_utterances": 3,
353
+ "sentence_error_rate": 0.5,
354
+ "exact_sentence_rate": 0.5,
355
+ "character_errors": 21,
356
+ "reference_characters": 397,
357
+ "character_error_rate": 0.05289672544080604,
358
+ "mean_utterance_wer": 0.09074074074074073,
359
+ "p95_utterance_wer": 0.2777777777777778,
360
+ "worst_utterance_wer": 0.3333333333333333
361
+ },
362
+ "narrative": {
363
+ "utterances": 24,
364
+ "semantic_errors": 4,
365
+ "semantic_words": 352,
366
+ "semantic_wer": 0.011363636363636364,
367
+ "failed_utterances": 3,
368
+ "sentence_error_rate": 0.125,
369
+ "exact_sentence_rate": 0.875,
370
+ "character_errors": 19,
371
+ "reference_characters": 1762,
372
+ "character_error_rate": 0.010783200908059024,
373
+ "mean_utterance_wer": 0.013123219373219372,
374
+ "p95_utterance_wer": 0.10194444444444431,
375
+ "worst_utterance_wer": 0.15384615384615385
376
+ },
377
+ "numbers": {
378
+ "utterances": 6,
379
+ "semantic_errors": 8,
380
+ "semantic_words": 64,
381
+ "semantic_wer": 0.125,
382
+ "failed_utterances": 3,
383
+ "sentence_error_rate": 0.5,
384
+ "exact_sentence_rate": 0.5,
385
+ "character_errors": 16,
386
+ "reference_characters": 322,
387
+ "character_error_rate": 0.049689440993788817,
388
+ "mean_utterance_wer": 0.1428030303030303,
389
+ "p95_utterance_wer": 0.35625,
390
+ "worst_utterance_wer": 0.375
391
+ },
392
+ "punctuation": {
393
+ "utterances": 6,
394
+ "semantic_errors": 0,
395
+ "semantic_words": 65,
396
+ "semantic_wer": 0.0,
397
+ "failed_utterances": 0,
398
+ "sentence_error_rate": 0.0,
399
+ "exact_sentence_rate": 1.0,
400
+ "character_errors": 0,
401
+ "reference_characters": 402,
402
+ "character_error_rate": 0.0,
403
+ "mean_utterance_wer": 0.0,
404
+ "p95_utterance_wer": 0.0,
405
+ "worst_utterance_wer": 0.0
406
+ },
407
+ "question": {
408
+ "utterances": 15,
409
+ "semantic_errors": 0,
410
+ "semantic_words": 165,
411
+ "semantic_wer": 0.0,
412
+ "failed_utterances": 0,
413
+ "sentence_error_rate": 0.0,
414
+ "exact_sentence_rate": 1.0,
415
+ "character_errors": 0,
416
+ "reference_characters": 834,
417
+ "character_error_rate": 0.0,
418
+ "mean_utterance_wer": 0.0,
419
+ "p95_utterance_wer": 0.0,
420
+ "worst_utterance_wer": 0.0
421
+ },
422
+ "rare_words": {
423
+ "utterances": 6,
424
+ "semantic_errors": 2,
425
+ "semantic_words": 57,
426
+ "semantic_wer": 0.03508771929824561,
427
+ "failed_utterances": 1,
428
+ "sentence_error_rate": 0.16666666666666666,
429
+ "exact_sentence_rate": 0.8333333333333334,
430
+ "character_errors": 1,
431
+ "reference_characters": 500,
432
+ "character_error_rate": 0.002,
433
+ "mean_utterance_wer": 0.030303030303030304,
434
+ "p95_utterance_wer": 0.13636363636363635,
435
+ "worst_utterance_wer": 0.18181818181818182
436
+ },
437
+ "short": {
438
+ "utterances": 24,
439
+ "semantic_errors": 5,
440
+ "semantic_words": 105,
441
+ "semantic_wer": 0.047619047619047616,
442
+ "failed_utterances": 4,
443
+ "sentence_error_rate": 0.16666666666666666,
444
+ "exact_sentence_rate": 0.8333333333333334,
445
+ "character_errors": 17,
446
+ "reference_characters": 458,
447
+ "character_error_rate": 0.03711790393013101,
448
+ "mean_utterance_wer": 0.051388888888888894,
449
+ "p95_utterance_wer": 0.32083333333333314,
450
+ "worst_utterance_wer": 0.4
451
+ },
452
+ "technical": {
453
+ "utterances": 21,
454
+ "semantic_errors": 8,
455
+ "semantic_words": 256,
456
+ "semantic_wer": 0.03125,
457
+ "failed_utterances": 4,
458
+ "sentence_error_rate": 0.19047619047619047,
459
+ "exact_sentence_rate": 0.8095238095238095,
460
+ "character_errors": 6,
461
+ "reference_characters": 1497,
462
+ "character_error_rate": 0.004008016032064128,
463
+ "mean_utterance_wer": 0.030455259026687595,
464
+ "p95_utterance_wer": 0.2,
465
+ "worst_utterance_wer": 0.2857142857142857
466
+ }
467
+ }
468
+ },
469
+ "kitten-nano-bruno": {
470
+ "overall": {
471
+ "utterances": 200,
472
+ "semantic_errors": 25,
473
+ "semantic_words": 2608,
474
+ "semantic_wer": 0.009585889570552147,
475
+ "failed_utterances": 15,
476
+ "sentence_error_rate": 0.075,
477
+ "exact_sentence_rate": 0.925,
478
+ "character_errors": 41,
479
+ "reference_characters": 13718,
480
+ "character_error_rate": 0.002988773873742528,
481
+ "mean_utterance_wer": 0.012458457692977818,
482
+ "p95_utterance_wer": 0.09136363636363622,
483
+ "worst_utterance_wer": 0.375
484
+ },
485
+ "categories": {
486
+ "conversational": {
487
+ "utterances": 25,
488
+ "semantic_errors": 1,
489
+ "semantic_words": 309,
490
+ "semantic_wer": 0.003236245954692557,
491
+ "failed_utterances": 1,
492
+ "sentence_error_rate": 0.04,
493
+ "exact_sentence_rate": 0.96,
494
+ "character_errors": 2,
495
+ "reference_characters": 1501,
496
+ "character_error_rate": 0.0013324450366422385,
497
+ "mean_utterance_wer": 0.002857142857142857,
498
+ "p95_utterance_wer": 0.0,
499
+ "worst_utterance_wer": 0.07142857142857142
500
+ },
501
+ "descriptive": {
502
+ "utterances": 16,
503
+ "semantic_errors": 0,
504
+ "semantic_words": 226,
505
+ "semantic_wer": 0.0,
506
+ "failed_utterances": 0,
507
+ "sentence_error_rate": 0.0,
508
+ "exact_sentence_rate": 1.0,
509
+ "character_errors": 0,
510
+ "reference_characters": 1146,
511
+ "character_error_rate": 0.0,
512
+ "mean_utterance_wer": 0.0,
513
+ "p95_utterance_wer": 0.0,
514
+ "worst_utterance_wer": 0.0
515
+ },
516
+ "emotional": {
517
+ "utterances": 24,
518
+ "semantic_errors": 0,
519
+ "semantic_words": 329,
520
+ "semantic_wer": 0.0,
521
+ "failed_utterances": 0,
522
+ "sentence_error_rate": 0.0,
523
+ "exact_sentence_rate": 1.0,
524
+ "character_errors": 0,
525
+ "reference_characters": 1574,
526
+ "character_error_rate": 0.0,
527
+ "mean_utterance_wer": 0.0,
528
+ "p95_utterance_wer": 0.0,
529
+ "worst_utterance_wer": 0.0
530
+ },
531
+ "homographs": {
532
+ "utterances": 6,
533
+ "semantic_errors": 1,
534
+ "semantic_words": 70,
535
+ "semantic_wer": 0.014285714285714285,
536
+ "failed_utterances": 1,
537
+ "sentence_error_rate": 0.16666666666666666,
538
+ "exact_sentence_rate": 0.8333333333333334,
539
+ "character_errors": 1,
540
+ "reference_characters": 375,
541
+ "character_error_rate": 0.0026666666666666666,
542
+ "mean_utterance_wer": 0.016666666666666666,
543
+ "p95_utterance_wer": 0.07500000000000001,
544
+ "worst_utterance_wer": 0.1
545
+ },
546
+ "long": {
547
+ "utterances": 15,
548
+ "semantic_errors": 1,
549
+ "semantic_words": 432,
550
+ "semantic_wer": 0.0023148148148148147,
551
+ "failed_utterances": 1,
552
+ "sentence_error_rate": 0.06666666666666667,
553
+ "exact_sentence_rate": 0.9333333333333333,
554
+ "character_errors": 4,
555
+ "reference_characters": 2206,
556
+ "character_error_rate": 0.001813236627379873,
557
+ "mean_utterance_wer": 0.002564102564102564,
558
+ "p95_utterance_wer": 0.011538461538461497,
559
+ "worst_utterance_wer": 0.038461538461538464
560
+ },
561
+ "long_range": {
562
+ "utterances": 6,
563
+ "semantic_errors": 0,
564
+ "semantic_words": 121,
565
+ "semantic_wer": 0.0,
566
+ "failed_utterances": 0,
567
+ "sentence_error_rate": 0.0,
568
+ "exact_sentence_rate": 1.0,
569
+ "character_errors": 0,
570
+ "reference_characters": 744,
571
+ "character_error_rate": 0.0,
572
+ "mean_utterance_wer": 0.0,
573
+ "p95_utterance_wer": 0.0,
574
+ "worst_utterance_wer": 0.0
575
+ },
576
+ "names_places": {
577
+ "utterances": 6,
578
+ "semantic_errors": 3,
579
+ "semantic_words": 57,
580
+ "semantic_wer": 0.05263157894736842,
581
+ "failed_utterances": 1,
582
+ "sentence_error_rate": 0.16666666666666666,
583
+ "exact_sentence_rate": 0.8333333333333334,
584
+ "character_errors": 12,
585
+ "reference_characters": 397,
586
+ "character_error_rate": 0.030226700251889168,
587
+ "mean_utterance_wer": 0.05555555555555555,
588
+ "p95_utterance_wer": 0.25,
589
+ "worst_utterance_wer": 0.3333333333333333
590
+ },
591
+ "narrative": {
592
+ "utterances": 24,
593
+ "semantic_errors": 4,
594
+ "semantic_words": 352,
595
+ "semantic_wer": 0.011363636363636364,
596
+ "failed_utterances": 3,
597
+ "sentence_error_rate": 0.125,
598
+ "exact_sentence_rate": 0.875,
599
+ "character_errors": 10,
600
+ "reference_characters": 1762,
601
+ "character_error_rate": 0.0056753688989784334,
602
+ "mean_utterance_wer": 0.009852296181630547,
603
+ "p95_utterance_wer": 0.05789473684210525,
604
+ "worst_utterance_wer": 0.125
605
+ },
606
+ "numbers": {
607
+ "utterances": 6,
608
+ "semantic_errors": 6,
609
+ "semantic_words": 64,
610
+ "semantic_wer": 0.09375,
611
+ "failed_utterances": 3,
612
+ "sentence_error_rate": 0.5,
613
+ "exact_sentence_rate": 0.5,
614
+ "character_errors": 5,
615
+ "reference_characters": 322,
616
+ "character_error_rate": 0.015527950310559006,
617
+ "mean_utterance_wer": 0.11098484848484848,
618
+ "p95_utterance_wer": 0.33125,
619
+ "worst_utterance_wer": 0.375
620
+ },
621
+ "punctuation": {
622
+ "utterances": 6,
623
+ "semantic_errors": 0,
624
+ "semantic_words": 65,
625
+ "semantic_wer": 0.0,
626
+ "failed_utterances": 0,
627
+ "sentence_error_rate": 0.0,
628
+ "exact_sentence_rate": 1.0,
629
+ "character_errors": 0,
630
+ "reference_characters": 402,
631
+ "character_error_rate": 0.0,
632
+ "mean_utterance_wer": 0.0,
633
+ "p95_utterance_wer": 0.0,
634
+ "worst_utterance_wer": 0.0
635
+ },
636
+ "question": {
637
+ "utterances": 15,
638
+ "semantic_errors": 0,
639
+ "semantic_words": 165,
640
+ "semantic_wer": 0.0,
641
+ "failed_utterances": 0,
642
+ "sentence_error_rate": 0.0,
643
+ "exact_sentence_rate": 1.0,
644
+ "character_errors": 0,
645
+ "reference_characters": 834,
646
+ "character_error_rate": 0.0,
647
+ "mean_utterance_wer": 0.0,
648
+ "p95_utterance_wer": 0.0,
649
+ "worst_utterance_wer": 0.0
650
+ },
651
+ "rare_words": {
652
+ "utterances": 6,
653
+ "semantic_errors": 2,
654
+ "semantic_words": 57,
655
+ "semantic_wer": 0.03508771929824561,
656
+ "failed_utterances": 1,
657
+ "sentence_error_rate": 0.16666666666666666,
658
+ "exact_sentence_rate": 0.8333333333333334,
659
+ "character_errors": 1,
660
+ "reference_characters": 500,
661
+ "character_error_rate": 0.002,
662
+ "mean_utterance_wer": 0.030303030303030304,
663
+ "p95_utterance_wer": 0.13636363636363635,
664
+ "worst_utterance_wer": 0.18181818181818182
665
+ },
666
+ "short": {
667
+ "utterances": 24,
668
+ "semantic_errors": 2,
669
+ "semantic_words": 105,
670
+ "semantic_wer": 0.01904761904761905,
671
+ "failed_utterances": 2,
672
+ "sentence_error_rate": 0.08333333333333333,
673
+ "exact_sentence_rate": 0.9166666666666666,
674
+ "character_errors": 3,
675
+ "reference_characters": 458,
676
+ "character_error_rate": 0.006550218340611353,
677
+ "mean_utterance_wer": 0.01875,
678
+ "p95_utterance_wer": 0.1699999999999996,
679
+ "worst_utterance_wer": 0.25
680
+ },
681
+ "technical": {
682
+ "utterances": 21,
683
+ "semantic_errors": 5,
684
+ "semantic_words": 256,
685
+ "semantic_wer": 0.01953125,
686
+ "failed_utterances": 2,
687
+ "sentence_error_rate": 0.09523809523809523,
688
+ "exact_sentence_rate": 0.9047619047619048,
689
+ "character_errors": 3,
690
+ "reference_characters": 1497,
691
+ "character_error_rate": 0.002004008016032064,
692
+ "mean_utterance_wer": 0.019727891156462583,
693
+ "p95_utterance_wer": 0.2,
694
+ "worst_utterance_wer": 0.21428571428571427
695
+ }
696
+ }
697
+ },
698
+ "piper-ryan-low": {
699
+ "overall": {
700
+ "utterances": 200,
701
+ "semantic_errors": 32,
702
+ "semantic_words": 2608,
703
+ "semantic_wer": 0.012269938650306749,
704
+ "failed_utterances": 19,
705
+ "sentence_error_rate": 0.095,
706
+ "exact_sentence_rate": 0.905,
707
+ "character_errors": 67,
708
+ "reference_characters": 13718,
709
+ "character_error_rate": 0.00488409389123779,
710
+ "mean_utterance_wer": 0.01766635309135309,
711
+ "p95_utterance_wer": 0.11464646464646343,
712
+ "worst_utterance_wer": 0.6
713
+ },
714
+ "categories": {
715
+ "conversational": {
716
+ "utterances": 25,
717
+ "semantic_errors": 2,
718
+ "semantic_words": 309,
719
+ "semantic_wer": 0.006472491909385114,
720
+ "failed_utterances": 1,
721
+ "sentence_error_rate": 0.04,
722
+ "exact_sentence_rate": 0.96,
723
+ "character_errors": 5,
724
+ "reference_characters": 1501,
725
+ "character_error_rate": 0.0033311125916055963,
726
+ "mean_utterance_wer": 0.008,
727
+ "p95_utterance_wer": 0.0,
728
+ "worst_utterance_wer": 0.2
729
+ },
730
+ "descriptive": {
731
+ "utterances": 16,
732
+ "semantic_errors": 2,
733
+ "semantic_words": 226,
734
+ "semantic_wer": 0.008849557522123894,
735
+ "failed_utterances": 2,
736
+ "sentence_error_rate": 0.125,
737
+ "exact_sentence_rate": 0.875,
738
+ "character_errors": 5,
739
+ "reference_characters": 1146,
740
+ "character_error_rate": 0.004363001745200698,
741
+ "mean_utterance_wer": 0.008630952380952382,
742
+ "p95_utterance_wer": 0.06785714285714287,
743
+ "worst_utterance_wer": 0.07142857142857142
744
+ },
745
+ "emotional": {
746
+ "utterances": 24,
747
+ "semantic_errors": 0,
748
+ "semantic_words": 329,
749
+ "semantic_wer": 0.0,
750
+ "failed_utterances": 0,
751
+ "sentence_error_rate": 0.0,
752
+ "exact_sentence_rate": 1.0,
753
+ "character_errors": 0,
754
+ "reference_characters": 1574,
755
+ "character_error_rate": 0.0,
756
+ "mean_utterance_wer": 0.0,
757
+ "p95_utterance_wer": 0.0,
758
+ "worst_utterance_wer": 0.0
759
+ },
760
+ "homographs": {
761
+ "utterances": 6,
762
+ "semantic_errors": 1,
763
+ "semantic_words": 70,
764
+ "semantic_wer": 0.014285714285714285,
765
+ "failed_utterances": 1,
766
+ "sentence_error_rate": 0.16666666666666666,
767
+ "exact_sentence_rate": 0.8333333333333334,
768
+ "character_errors": 1,
769
+ "reference_characters": 375,
770
+ "character_error_rate": 0.0026666666666666666,
771
+ "mean_utterance_wer": 0.016666666666666666,
772
+ "p95_utterance_wer": 0.07500000000000001,
773
+ "worst_utterance_wer": 0.1
774
+ },
775
+ "long": {
776
+ "utterances": 15,
777
+ "semantic_errors": 2,
778
+ "semantic_words": 432,
779
+ "semantic_wer": 0.004629629629629629,
780
+ "failed_utterances": 2,
781
+ "sentence_error_rate": 0.13333333333333333,
782
+ "exact_sentence_rate": 0.8666666666666667,
783
+ "character_errors": 9,
784
+ "reference_characters": 2206,
785
+ "character_error_rate": 0.004079782411604714,
786
+ "mean_utterance_wer": 0.0052307692307692315,
787
+ "p95_utterance_wer": 0.03892307692307692,
788
+ "worst_utterance_wer": 0.04
789
+ },
790
+ "long_range": {
791
+ "utterances": 6,
792
+ "semantic_errors": 0,
793
+ "semantic_words": 121,
794
+ "semantic_wer": 0.0,
795
+ "failed_utterances": 0,
796
+ "sentence_error_rate": 0.0,
797
+ "exact_sentence_rate": 1.0,
798
+ "character_errors": 0,
799
+ "reference_characters": 744,
800
+ "character_error_rate": 0.0,
801
+ "mean_utterance_wer": 0.0,
802
+ "p95_utterance_wer": 0.0,
803
+ "worst_utterance_wer": 0.0
804
+ },
805
+ "names_places": {
806
+ "utterances": 6,
807
+ "semantic_errors": 4,
808
+ "semantic_words": 57,
809
+ "semantic_wer": 0.07017543859649122,
810
+ "failed_utterances": 2,
811
+ "sentence_error_rate": 0.3333333333333333,
812
+ "exact_sentence_rate": 0.6666666666666667,
813
+ "character_errors": 19,
814
+ "reference_characters": 397,
815
+ "character_error_rate": 0.04785894206549118,
816
+ "mean_utterance_wer": 0.07407407407407407,
817
+ "p95_utterance_wer": 0.2777777777777778,
818
+ "worst_utterance_wer": 0.3333333333333333
819
+ },
820
+ "narrative": {
821
+ "utterances": 24,
822
+ "semantic_errors": 0,
823
+ "semantic_words": 352,
824
+ "semantic_wer": 0.0,
825
+ "failed_utterances": 0,
826
+ "sentence_error_rate": 0.0,
827
+ "exact_sentence_rate": 1.0,
828
+ "character_errors": 0,
829
+ "reference_characters": 1762,
830
+ "character_error_rate": 0.0,
831
+ "mean_utterance_wer": 0.0,
832
+ "p95_utterance_wer": 0.0,
833
+ "worst_utterance_wer": 0.0
834
+ },
835
+ "numbers": {
836
+ "utterances": 6,
837
+ "semantic_errors": 5,
838
+ "semantic_words": 64,
839
+ "semantic_wer": 0.078125,
840
+ "failed_utterances": 2,
841
+ "sentence_error_rate": 0.3333333333333333,
842
+ "exact_sentence_rate": 0.6666666666666667,
843
+ "character_errors": 3,
844
+ "reference_characters": 322,
845
+ "character_error_rate": 0.009316770186335404,
846
+ "mean_utterance_wer": 0.09583333333333333,
847
+ "p95_utterance_wer": 0.33125,
848
+ "worst_utterance_wer": 0.375
849
+ },
850
+ "punctuation": {
851
+ "utterances": 6,
852
+ "semantic_errors": 0,
853
+ "semantic_words": 65,
854
+ "semantic_wer": 0.0,
855
+ "failed_utterances": 0,
856
+ "sentence_error_rate": 0.0,
857
+ "exact_sentence_rate": 1.0,
858
+ "character_errors": 0,
859
+ "reference_characters": 402,
860
+ "character_error_rate": 0.0,
861
+ "mean_utterance_wer": 0.0,
862
+ "p95_utterance_wer": 0.0,
863
+ "worst_utterance_wer": 0.0
864
+ },
865
+ "question": {
866
+ "utterances": 15,
867
+ "semantic_errors": 0,
868
+ "semantic_words": 165,
869
+ "semantic_wer": 0.0,
870
+ "failed_utterances": 0,
871
+ "sentence_error_rate": 0.0,
872
+ "exact_sentence_rate": 1.0,
873
+ "character_errors": 0,
874
+ "reference_characters": 834,
875
+ "character_error_rate": 0.0,
876
+ "mean_utterance_wer": 0.0,
877
+ "p95_utterance_wer": 0.0,
878
+ "worst_utterance_wer": 0.0
879
+ },
880
+ "rare_words": {
881
+ "utterances": 6,
882
+ "semantic_errors": 5,
883
+ "semantic_words": 57,
884
+ "semantic_wer": 0.08771929824561403,
885
+ "failed_utterances": 3,
886
+ "sentence_error_rate": 0.5,
887
+ "exact_sentence_rate": 0.5,
888
+ "character_errors": 9,
889
+ "reference_characters": 500,
890
+ "character_error_rate": 0.018,
891
+ "mean_utterance_wer": 0.0803030303030303,
892
+ "p95_utterance_wer": 0.1954545454545455,
893
+ "worst_utterance_wer": 0.2
894
+ },
895
+ "short": {
896
+ "utterances": 24,
897
+ "semantic_errors": 4,
898
+ "semantic_words": 105,
899
+ "semantic_wer": 0.0380952380952381,
900
+ "failed_utterances": 2,
901
+ "sentence_error_rate": 0.08333333333333333,
902
+ "exact_sentence_rate": 0.9166666666666666,
903
+ "character_errors": 10,
904
+ "reference_characters": 458,
905
+ "character_error_rate": 0.021834061135371178,
906
+ "mean_utterance_wer": 0.03888888888888889,
907
+ "p95_utterance_wer": 0.2833333333333326,
908
+ "worst_utterance_wer": 0.6
909
+ },
910
+ "technical": {
911
+ "utterances": 21,
912
+ "semantic_errors": 7,
913
+ "semantic_words": 256,
914
+ "semantic_wer": 0.02734375,
915
+ "failed_utterances": 4,
916
+ "sentence_error_rate": 0.19047619047619047,
917
+ "exact_sentence_rate": 0.8095238095238095,
918
+ "character_errors": 6,
919
+ "reference_characters": 1497,
920
+ "character_error_rate": 0.004008016032064128,
921
+ "mean_utterance_wer": 0.027719899148470575,
922
+ "p95_utterance_wer": 0.2,
923
+ "worst_utterance_wer": 0.21428571428571427
924
+ }
925
+ }
926
+ }
927
+ }
928
+ }