Add Silero VAD 6.2.3 Core ML model

README.md

@@ -129,10 +129,11 @@ For offline setups, download this repository and set `SCRIPTRS_MODELS_DIR` to it
 | `parakeet-v2/model.safetensors` | CUDA | nvidia/parakeet-tdt-0.6b-v2, exported from the NeMo checkpoint |
 | `parakeet-v3/model.safetensors` | CUDA | nvidia/parakeet-tdt-0.6b-v3, exported from the NeMo checkpoint |
 | `parakeet-v2/vocab.txt`, `parakeet-v3/parakeet_vocab.json` | both | NVIDIA tokenizers |
-| `vad/silero-vad.mlmodelc` | Core ML | [Silero VAD](https://github.com/snakers4/silero-vad) |
+| `vad/silero-vad.mlmodelc` | Core ML | [Silero VAD](https://github.com/snakers4/silero-vad) 5.1.2, via [aufklarer/Silero-VAD-v5-CoreML](https://huggingface.co/aufklarer/Silero-VAD-v5-CoreML) |
+| `vad/silero-vad-v6.mlmodelc` | Core ML | Silero VAD 6.2.3, via [aufklarer/Silero-VAD-v6.2.1-CoreML](https://huggingface.co/aufklarer/Silero-VAD-v6.2.1-CoreML), which has the same weights |
 | `vad/silero-vad-v5.safetensors` | CUDA | Silero VAD 5.1.2, 16 kHz branch |
 
-The native CUDA files also live in model-named repositories, which scriptrs 0.3.1 and later download from: [avencera/parakeet-tdt-0.6b-v2-cuda](https://huggingface.co/avencera/parakeet-tdt-0.6b-v2-cuda), [avencera/parakeet-tdt-0.6b-v3-cuda](https://huggingface.co/avencera/parakeet-tdt-0.6b-v3-cuda) and [avencera/silero-vad-v5-safetensors](https://huggingface.co/avencera/silero-vad-v5-safetensors). The copies here serve scriptrs 0.3.0.
+The native CUDA files also live in model-named repositories, which scriptrs 0.3.1 and later download from: [avencera/parakeet-tdt-0.6b-v2-cuda](https://huggingface.co/avencera/parakeet-tdt-0.6b-v2-cuda), [avencera/parakeet-tdt-0.6b-v3-cuda](https://huggingface.co/avencera/parakeet-tdt-0.6b-v3-cuda) and [avencera/silero-vad-v5-safetensors](https://huggingface.co/avencera/silero-vad-v5-safetensors), and from scriptrs 0.5.0 also [avencera/silero-vad-v6-safetensors](https://huggingface.co/avencera/silero-vad-v6-safetensors). The copies here serve scriptrs 0.3.0.
 
 The native weight files record their source and its SHA-256 in the safetensors metadata. `PROVENANCE.json` and `PROVENANCE-native.json` list sources, revisions and checksums.
 

vad/silero-vad-v6.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b777c3751d72b7430eac7f8544769a3d918faf77c15db184fec30e44c56007a3
+size 243

vad/silero-vad-v6.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6fcd92c3132c9c718e5f54e0e770a8c8075beaa50a5b212a6287273b4ddae67
+size 399

vad/silero-vad-v6.mlmodelc/metadata.json

@@ -0,0 +1,112 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float16",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "probability",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 128]",
+        "name" : "h_out",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 128]",
+        "name" : "c_out",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios17.sqrt" : 1,
+      "Ios17.transpose" : 1,
+      "Ios17.conv" : 6,
+      "Ios16.sigmoid" : 1,
+      "Ios17.sliceByIndex" : 2,
+      "Ios17.add" : 2,
+      "Ios17.lstm" : 1,
+      "Ios16.relu" : 5,
+      "Ios17.pow" : 2,
+      "Ios17.expandDims" : 3,
+      "Ios17.squeeze" : 3,
+      "Ios17.reshape" : 1,
+      "Pad" : 1
+    },
+    "computePrecision" : "Mixed (Float16, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.source_dialect" : "TorchScript",
+      "com.github.apple.coremltools.version" : "8.3.0",
+      "com.github.apple.coremltools.source" : "torch==2.9.1"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 576)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 576]",
+        "name" : "audio",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 128]",
+        "name" : "h",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 128]",
+        "name" : "c",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "silero_vad",
+    "method" : "predict"
+  }
+]

vad/silero-vad-v6.mlmodelc/model.mil

@@ -0,0 +1,104 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-torch", "2.9.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})]
+{
+    func main<ios17>(tensor<fp16, [1, 1, 576]> audio, tensor<fp16, [1, 1, 128]> c, tensor<fp16, [1, 1, 128]> h) {
+            tensor<int32, [6]> input_1_pad_0 = const()[name = tensor<string, []>("input_1_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 64])];
+            tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("reflect")];
+            tensor<fp16, []> const_0_to_fp16 = const()[name = tensor<string, []>("const_0_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1, 640]> input_1_cast_fp16 = pad(constant_val = const_0_to_fp16, mode = input_1_mode_0, pad = input_1_pad_0, x = audio)[name = tensor<string, []>("input_1_cast_fp16")];
+            tensor<string, []> x_1_pad_type_0 = const()[name = tensor<string, []>("x_1_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> x_1_strides_0 = const()[name = tensor<string, []>("x_1_strides_0"), val = tensor<int32, [1]>([128])];
+            tensor<int32, [2]> x_1_pad_0 = const()[name = tensor<string, []>("x_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_1_dilations_0 = const()[name = tensor<string, []>("x_1_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> x_1_groups_0 = const()[name = tensor<string, []>("x_1_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [258, 1, 256]> stft_conv_weight_to_fp16 = const()[name = tensor<string, []>("stft_conv_weight_to_fp16"), val = tensor<fp16, [258, 1, 256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [1, 258, 4]> x_1_cast_fp16 = conv(dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = stft_conv_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("x_1_cast_fp16")];
+            tensor<int32, [3]> var_42_begin_0 = const()[name = tensor<string, []>("op_42_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
+            tensor<int32, [3]> var_42_end_0 = const()[name = tensor<string, []>("op_42_end_0"), val = tensor<int32, [3]>([1, 129, 4])];
+            tensor<bool, [3]> var_42_end_mask_0 = const()[name = tensor<string, []>("op_42_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
+            tensor<fp16, [1, 129, 4]> var_42_cast_fp16 = slice_by_index(begin = var_42_begin_0, end = var_42_end_0, end_mask = var_42_end_mask_0, x = x_1_cast_fp16)[name = tensor<string, []>("op_42_cast_fp16")];
+            tensor<int32, [3]> var_57_begin_0 = const()[name = tensor<string, []>("op_57_begin_0"), val = tensor<int32, [3]>([0, 129, 0])];
+            tensor<int32, [3]> var_57_end_0 = const()[name = tensor<string, []>("op_57_end_0"), val = tensor<int32, [3]>([1, 258, 4])];
+            tensor<bool, [3]> var_57_end_mask_0 = const()[name = tensor<string, []>("op_57_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
+            tensor<fp16, [1, 129, 4]> var_57_cast_fp16 = slice_by_index(begin = var_57_begin_0, end = var_57_end_0, end_mask = var_57_end_mask_0, x = x_1_cast_fp16)[name = tensor<string, []>("op_57_cast_fp16")];
+            tensor<fp16, []> var_63_promoted_to_fp16 = const()[name = tensor<string, []>("op_63_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
+            tensor<fp16, [1, 129, 4]> var_64_cast_fp16 = pow(x = var_42_cast_fp16, y = var_63_promoted_to_fp16)[name = tensor<string, []>("op_64_cast_fp16")];
+            tensor<fp16, []> var_65_promoted_to_fp16 = const()[name = tensor<string, []>("op_65_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
+            tensor<fp16, [1, 129, 4]> var_66_cast_fp16 = pow(x = var_57_cast_fp16, y = var_65_promoted_to_fp16)[name = tensor<string, []>("op_66_cast_fp16")];
+            tensor<fp16, [1, 129, 4]> var_68_cast_fp16 = add(x = var_64_cast_fp16, y = var_66_cast_fp16)[name = tensor<string, []>("op_68_cast_fp16")];
+            tensor<fp16, []> var_70_to_fp16 = const()[name = tensor<string, []>("op_70_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
+            tensor<fp16, [1, 129, 4]> var_71_cast_fp16 = add(x = var_68_cast_fp16, y = var_70_to_fp16)[name = tensor<string, []>("op_71_cast_fp16")];
+            tensor<fp16, [1, 129, 4]> input_3_cast_fp16 = sqrt(x = var_71_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
+            tensor<string, []> input_5_pad_type_0 = const()[name = tensor<string, []>("input_5_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_5_pad_0 = const()[name = tensor<string, []>("input_5_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_5_strides_0 = const()[name = tensor<string, []>("input_5_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_5_dilations_0 = const()[name = tensor<string, []>("input_5_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_5_groups_0 = const()[name = tensor<string, []>("input_5_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [128, 129, 3]> encoder_0_weight_to_fp16 = const()[name = tensor<string, []>("encoder_0_weight_to_fp16"), val = tensor<fp16, [128, 129, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132224)))];
+            tensor<fp16, [128]> encoder_0_bias_to_fp16 = const()[name = tensor<string, []>("encoder_0_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231360)))];
+            tensor<fp16, [1, 128, 4]> input_5_cast_fp16 = conv(bias = encoder_0_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = encoder_0_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
+            tensor<fp16, [1, 128, 4]> input_7_cast_fp16 = relu(x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
+            tensor<string, []> input_9_pad_type_0 = const()[name = tensor<string, []>("input_9_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_9_pad_0 = const()[name = tensor<string, []>("input_9_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_9_strides_0 = const()[name = tensor<string, []>("input_9_strides_0"), val = tensor<int32, [1]>([2])];
+            tensor<int32, [1]> input_9_dilations_0 = const()[name = tensor<string, []>("input_9_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_9_groups_0 = const()[name = tensor<string, []>("input_9_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [64, 128, 3]> encoder_1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_1_weight_to_fp16"), val = tensor<fp16, [64, 128, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231680)))];
+            tensor<fp16, [64]> encoder_1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_1_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(280896)))];
+            tensor<fp16, [1, 64, 2]> input_9_cast_fp16 = conv(bias = encoder_1_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = encoder_1_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
+            tensor<fp16, [1, 64, 2]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
+            tensor<string, []> input_13_pad_type_0 = const()[name = tensor<string, []>("input_13_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_13_pad_0 = const()[name = tensor<string, []>("input_13_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_13_strides_0 = const()[name = tensor<string, []>("input_13_strides_0"), val = tensor<int32, [1]>([2])];
+            tensor<int32, [1]> input_13_dilations_0 = const()[name = tensor<string, []>("input_13_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_13_groups_0 = const()[name = tensor<string, []>("input_13_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [64, 64, 3]> encoder_2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_2_weight_to_fp16"), val = tensor<fp16, [64, 64, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281088)))];
+            tensor<fp16, [64]> encoder_2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_2_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(305728)))];
+            tensor<fp16, [1, 64, 1]> input_13_cast_fp16 = conv(bias = encoder_2_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = encoder_2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
+            tensor<fp16, [1, 64, 1]> input_15_cast_fp16 = relu(x = input_13_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
+            tensor<string, []> input_17_pad_type_0 = const()[name = tensor<string, []>("input_17_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_17_pad_0 = const()[name = tensor<string, []>("input_17_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [1]> input_17_strides_0 = const()[name = tensor<string, []>("input_17_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> input_17_dilations_0 = const()[name = tensor<string, []>("input_17_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> input_17_groups_0 = const()[name = tensor<string, []>("input_17_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [128, 64, 3]> encoder_3_weight_to_fp16 = const()[name = tensor<string, []>("encoder_3_weight_to_fp16"), val = tensor<fp16, [128, 64, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(305920)))];
+            tensor<fp16, [128]> encoder_3_bias_to_fp16 = const()[name = tensor<string, []>("encoder_3_bias_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355136)))];
+            tensor<fp16, [1, 128, 1]> input_17_cast_fp16 = conv(bias = encoder_3_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = encoder_3_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
+            tensor<fp16, [1, 128, 1]> x_cast_fp16 = relu(x = input_17_cast_fp16)[name = tensor<string, []>("x_cast_fp16")];
+            tensor<int32, [3]> transpose_0_perm_0 = const()[name = tensor<string, []>("transpose_0_perm_0"), val = tensor<int32, [3]>([2, 0, 1])];
+            tensor<int32, [1]> var_138_batch_first_lstm_h0_squeeze_axes_0 = const()[name = tensor<string, []>("op_138_batch_first_lstm_h0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 128]> var_138_batch_first_lstm_h0_squeeze_cast_fp16 = squeeze(axes = var_138_batch_first_lstm_h0_squeeze_axes_0, x = h)[name = tensor<string, []>("op_138_batch_first_lstm_h0_squeeze_cast_fp16")];
+            tensor<int32, [1]> var_138_batch_first_lstm_c0_squeeze_axes_0 = const()[name = tensor<string, []>("op_138_batch_first_lstm_c0_squeeze_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 128]> var_138_batch_first_lstm_c0_squeeze_cast_fp16 = squeeze(axes = var_138_batch_first_lstm_c0_squeeze_axes_0, x = c)[name = tensor<string, []>("op_138_batch_first_lstm_c0_squeeze_cast_fp16")];
+            tensor<string, []> var_138_batch_first_direction_0 = const()[name = tensor<string, []>("op_138_batch_first_direction_0"), val = tensor<string, []>("forward")];
+            tensor<bool, []> var_138_batch_first_output_sequence_0 = const()[name = tensor<string, []>("op_138_batch_first_output_sequence_0"), val = tensor<bool, []>(true)];
+            tensor<string, []> var_138_batch_first_recurrent_activation_0 = const()[name = tensor<string, []>("op_138_batch_first_recurrent_activation_0"), val = tensor<string, []>("sigmoid")];
+            tensor<string, []> var_138_batch_first_cell_activation_0 = const()[name = tensor<string, []>("op_138_batch_first_cell_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<string, []> var_138_batch_first_activation_0 = const()[name = tensor<string, []>("op_138_batch_first_activation_0"), val = tensor<string, []>("tanh")];
+            tensor<fp16, [512, 128]> concat_1_to_fp16 = const()[name = tensor<string, []>("concat_1_to_fp16"), val = tensor<fp16, [512, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355456)))];
+            tensor<fp16, [512, 128]> concat_2_to_fp16 = const()[name = tensor<string, []>("concat_2_to_fp16"), val = tensor<fp16, [512, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486592)))];
+            tensor<fp16, [512]> concat_0_to_fp16 = const()[name = tensor<string, []>("concat_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617728)))];
+            tensor<fp16, [1, 1, 128]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = x_cast_fp16)[name = tensor<string, []>("transpose_1")];
+            tensor<fp16, [1, 1, 128]> var_138_batch_first_cast_fp16_0, tensor<fp16, [1, 128]> var_138_batch_first_cast_fp16_1, tensor<fp16, [1, 128]> var_138_batch_first_cast_fp16_2 = lstm(activation = var_138_batch_first_activation_0, bias = concat_0_to_fp16, cell_activation = var_138_batch_first_cell_activation_0, direction = var_138_batch_first_direction_0, initial_c = var_138_batch_first_lstm_c0_squeeze_cast_fp16, initial_h = var_138_batch_first_lstm_h0_squeeze_cast_fp16, output_sequence = var_138_batch_first_output_sequence_0, recurrent_activation = var_138_batch_first_recurrent_activation_0, weight_hh = concat_2_to_fp16, weight_ih = concat_1_to_fp16, x = transpose_0_cast_fp16)[name = tensor<string, []>("op_138_batch_first_cast_fp16")];
+            tensor<int32, [1]> h_out_axes_0 = const()[name = tensor<string, []>("h_out_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 1, 128]> h_out = expand_dims(axes = h_out_axes_0, x = var_138_batch_first_cast_fp16_1)[name = tensor<string, []>("h_out_cast_fp16")];
+            tensor<int32, [1]> var_140_axes_0 = const()[name = tensor<string, []>("op_140_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 1, 128]> c_out = expand_dims(axes = var_140_axes_0, x = var_138_batch_first_cast_fp16_2)[name = tensor<string, []>("op_140_cast_fp16")];
+            tensor<int32, [1]> var_145_axes_0 = const()[name = tensor<string, []>("op_145_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp16, [1, 128]> var_145_cast_fp16 = squeeze(axes = var_145_axes_0, x = h_out)[name = tensor<string, []>("op_145_cast_fp16")];
+            tensor<int32, [1]> input_21_axes_0 = const()[name = tensor<string, []>("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1, 128, 1]> input_21_cast_fp16 = expand_dims(axes = input_21_axes_0, x = var_145_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
+            tensor<fp16, [1, 128, 1]> input_cast_fp16 = relu(x = input_21_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
+            tensor<string, []> var_159_pad_type_0 = const()[name = tensor<string, []>("op_159_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [1]> var_159_strides_0 = const()[name = tensor<string, []>("op_159_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> var_159_pad_0 = const()[name = tensor<string, []>("op_159_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> var_159_dilations_0 = const()[name = tensor<string, []>("op_159_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> var_159_groups_0 = const()[name = tensor<string, []>("op_159_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 128, 1]> decoder_conv_weight_to_fp16 = const()[name = tensor<string, []>("decoder_conv_weight_to_fp16"), val = tensor<fp16, [1, 128, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618816)))];
+            tensor<fp16, [1]> decoder_conv_bias_to_fp16 = const()[name = tensor<string, []>("decoder_conv_bias_to_fp16"), val = tensor<fp16, [1]>([-0x1.3fcp-1])];
+            tensor<fp16, [1, 1, 1]> var_159_cast_fp16 = conv(bias = decoder_conv_bias_to_fp16, dilations = var_159_dilations_0, groups = var_159_groups_0, pad = var_159_pad_0, pad_type = var_159_pad_type_0, strides = var_159_strides_0, weight = decoder_conv_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("op_159_cast_fp16")];
+            tensor<fp16, [1, 1, 1]> d_cast_fp16 = sigmoid(x = var_159_cast_fp16)[name = tensor<string, []>("d_cast_fp16")];
+            tensor<int32, [1]> var_162 = const()[name = tensor<string, []>("op_162"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1]> probability = reshape(shape = var_162, x = d_cast_fp16)[name = tensor<string, []>("op_163_cast_fp16")];
+        } -> (probability, h_out, c_out);
+}

vad/silero-vad-v6.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83210545de90c65195e8d6db1b349b7e5c31f989f48d0a908a8dc0e2f586e5f9
+size 619136