Upload 10 files

Embedding.mlmodelc/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef4e9c40ab61e1f3c1f04e05133ff2cedb385d75207823c15fdcce82f8a296a1
+oid sha256:9aed2af67ed99bcf0cbd9b4857563f3888128af655f99d37f8d2fcd000a269d7
 size 614

FBank.mlmodelc/analytics/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a57eb96c61ea3470b3c509eafd8085e3a05d023219096414554700bc31e3f25
+oid sha256:1b656c87aca7bc098bdc6569cd7d5e09fbe2ae7d0852485b148123dd71699dde
 size 243

FBank.mlmodelc/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33b0f747fa880e3446f676078fd1b33f7333db5e94cb956b321c24f14b2c12bf
-size 538
+oid sha256:adbafb83e071a43b26a8203b3ab6a30cee479e872f29ab420a8ca06fc3d50a54
+size 547

FBank.mlmodelc/metadata.json

@@ -1,6 +1,6 @@
 [
   {
-    "shortDescription" : "pyannote community-1 FBANK front-end (10 s audio -> 998 frames x 80 bins)",
+    "shortDescription" : "pyannote community-1 FBANK front-end (10 s audio -> channel-first 80x998 features)",
     "metadataOutputVersion" : "3.0",
     "outputSchema" : [
       {
@@ -24,7 +24,7 @@
     "license" : "CC-BY-4.0",
     "mlProgramOperationTypeHistogram" : {
       "Ios17.mul" : 2,
-      "Ios17.transpose" : 1,
+      "Ios17.transpose" : 2,
       "Ios17.sub" : 2,
       "Ios17.conv" : 4,
       "Ios17.log" : 1,
@@ -33,7 +33,7 @@
       "Ios17.add" : 1,
       "Ios17.clip" : 1,
       "Ios17.pow" : 2,
-      "Ios17.expandDims" : 3,
+      "Ios17.expandDims" : 4,
       "Ios17.squeeze" : 4,
       "Ios17.reshape" : 2,
       "Ios17.cast" : 6,

FBank.mlmodelc/model.mil

@@ -13,7 +13,7 @@ program(1.0)
             tensor<fp16, [?, 400, 998]> frames_1_cast_fp16 = conv(dilations = frames_1_dilations_0, groups = frames_1_groups_0, pad = frames_1_pad_0, pad_type = frames_1_pad_type_0, strides = frames_1_strides_0, weight = frame_kernel_to_fp16, x = audio_to_fp16)[name = tensor<string, []>("frames_1_cast_fp16")];
             tensor<int32, [3]> frames_3_perm_0 = const()[name = tensor<string, []>("frames_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
             tensor<int32, [2]> concat_0x = const()[name = tensor<string, []>("concat_0x"), val = tensor<int32, [2]>([-1, 400])];
-            tensor<fp16, [?, 998, 400]> frames_3_cast_fp16 = transpose(perm = frames_3_perm_0, x = frames_1_cast_fp16)[name = tensor<string, []>("transpose_0")];
+            tensor<fp16, [?, 998, 400]> frames_3_cast_fp16 = transpose(perm = frames_3_perm_0, x = frames_1_cast_fp16)[name = tensor<string, []>("transpose_1")];
             tensor<fp16, [?, 400]> frames_5_cast_fp16 = reshape(shape = concat_0x, x = frames_3_cast_fp16)[name = tensor<string, []>("frames_5_cast_fp16")];
             tensor<string, []> frames_5_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("frames_5_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
             tensor<int32, [1]> var_50_axes_0 = const()[name = tensor<string, []>("op_50_axes_0"), val = tensor<int32, [1]>([1])];
@@ -84,14 +84,18 @@ program(1.0)
             tensor<fp16, []> const_2_to_fp16 = const()[name = tensor<string, []>("const_2_to_fp16"), val = tensor<fp16, []>(inf)];
             tensor<fp16, [?, 80]> clip_0_cast_fp16 = clip(alpha = eps_to_fp16, beta = const_2_to_fp16, x = mel_1_cast_fp16)[name = tensor<string, []>("clip_0_cast_fp16")];
             tensor<string, []> clip_0_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("clip_0_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
-            tensor<fp32, []> mel_epsilon_0 = const()[name = tensor<string, []>("mel_epsilon_0"), val = tensor<fp32, []>(0x1p-149)];
+            tensor<fp32, []> mel_3_epsilon_0 = const()[name = tensor<string, []>("mel_3_epsilon_0"), val = tensor<fp32, []>(0x1p-149)];
             tensor<fp32, [?, 80]> clip_0_cast_fp16_to_fp32 = cast(dtype = clip_0_cast_fp16_to_fp32_dtype_0, x = clip_0_cast_fp16)[name = tensor<string, []>("cast_4")];
-            tensor<fp32, [?, 80]> mel = log(epsilon = mel_epsilon_0, x = clip_0_cast_fp16_to_fp32)[name = tensor<string, []>("mel")];
+            tensor<fp32, [?, 80]> mel_3 = log(epsilon = mel_3_epsilon_0, x = clip_0_cast_fp16_to_fp32)[name = tensor<string, []>("mel_3")];
             tensor<int32, [3]> concat_1x = const()[name = tensor<string, []>("concat_1x"), val = tensor<int32, [3]>([-1, 998, 80])];
-            tensor<string, []> mel_to_fp16_dtype_0 = const()[name = tensor<string, []>("mel_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
-            tensor<fp16, [?, 80]> mel_to_fp16 = cast(dtype = mel_to_fp16_dtype_0, x = mel)[name = tensor<string, []>("cast_3")];
-            tensor<fp16, [?, 998, 80]> var_157_cast_fp16 = reshape(shape = concat_1x, x = mel_to_fp16)[name = tensor<string, []>("op_157_cast_fp16")];
-            tensor<string, []> var_157_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_157_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
-            tensor<fp32, [?, 998, 80]> fbank = cast(dtype = var_157_cast_fp16_to_fp32_dtype_0, x = var_157_cast_fp16)[name = tensor<string, []>("cast_2")];
+            tensor<string, []> mel_3_to_fp16_dtype_0 = const()[name = tensor<string, []>("mel_3_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [?, 80]> mel_3_to_fp16 = cast(dtype = mel_3_to_fp16_dtype_0, x = mel_3)[name = tensor<string, []>("cast_3")];
+            tensor<fp16, [?, 998, 80]> mel_cast_fp16 = reshape(shape = concat_1x, x = mel_3_to_fp16)[name = tensor<string, []>("mel_cast_fp16")];
+            tensor<int32, [3]> var_161 = const()[name = tensor<string, []>("op_161"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_164_axes_0 = const()[name = tensor<string, []>("op_164_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [?, 80, 998]> var_162_cast_fp16 = transpose(perm = var_161, x = mel_cast_fp16)[name = tensor<string, []>("transpose_0")];
+            tensor<fp16, [?, 1, 80, 998]> var_164_cast_fp16 = expand_dims(axes = var_164_axes_0, x = var_162_cast_fp16)[name = tensor<string, []>("op_164_cast_fp16")];
+            tensor<string, []> var_164_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_164_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, [?, 1, 80, 998]> fbank = cast(dtype = var_164_cast_fp16_to_fp32_dtype_0, x = var_164_cast_fp16)[name = tensor<string, []>("cast_2")];
         } -> (fbank);
 }