Upload 29 files

Pipeline_Head_Fixed.mlmodelc/analytics/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5980b0b5b6afd629201028acd9d30ef139405a4ff8e3197551b5749757e19808
+oid sha256:cb4a6f09bdbefee13e23f4e5ebd7eca61ecf9cf1ec4e50ef6455aab8989120b5
 size 243

Pipeline_Head_Fixed.mlmodelc/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4d3a809042f1aafc6410902c356ad226e4104b9f92f21a266b85a89d501c8e3c
-size 505
+oid sha256:199004113e953fa92ba5ee1d208594a17e3d1a48b4289ae358b58ac78922ed26
+size 585

Pipeline_Head_Fixed.mlmodelc/metadata.json

@@ -0,0 +1,134 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float16",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 168 × 4)",
+        "shortDescription" : "",
+        "shape" : "[1, 168, 4]",
+        "name" : "speaker_preds",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 8 × 512)",
+        "shortDescription" : "",
+        "shape" : "[1, 8, 512]",
+        "name" : "chunk_pre_encoder_embs",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "chunk_pre_encoder_lengths",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 7,
+    "mlProgramOperationTypeHistogram" : {
+      "Transpose" : 192,
+      "Identity" : 2,
+      "Ios16.softmax" : 35,
+      "Split" : 17,
+      "Ios16.linear" : 247,
+      "Select" : 51,
+      "Ios16.add" : 173,
+      "Tile" : 1,
+      "Ios16.sigmoid" : 18,
+      "Ios16.logicalAnd" : 2,
+      "Pad" : 34,
+      "ExpandDims" : 8,
+      "Ios16.sub" : 1,
+      "Ios16.cast" : 4,
+      "Ios16.less" : 2,
+      "Ios16.conv" : 51,
+      "Ios16.matmul" : 87,
+      "Ios16.layerNorm" : 121,
+      "SliceByIndex" : 34,
+      "Ios16.reshape" : 174,
+      "Ios16.silu" : 51,
+      "Ios16.relu" : 20,
+      "Ios16.mul" : 107,
+      "Ios16.logicalNot" : 2
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "13.0",
+      "tvOS" : "16.0",
+      "visionOS" : "1.0",
+      "watchOS" : "9.0",
+      "iOS" : "16.0",
+      "macCatalyst" : "16.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-12-25",
+      "com.github.apple.coremltools.source" : "torch==2.9.1",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 168 × 512)",
+        "shortDescription" : "",
+        "shape" : "[1, 168, 512]",
+        "name" : "pre_encoder_embs",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "pre_encoder_lengths",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 8 × 512)",
+        "shortDescription" : "",
+        "shape" : "[1, 8, 512]",
+        "name" : "chunk_pre_encoder_embs",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "chunk_pre_encoder_lengths",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "Pipeline_Head_Fixed",
+    "method" : "predict"
+  }
+]

Pipeline_Head_Fixed.mlmodelc/weights/weight.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be56312ef2dbf57706aab7395fc2d5601ac6fbe6c553fd2b25069eba8da9b3b2
-size 235580992
+oid sha256:58897124dc41ee376bd757521ba368cdcd85249e5f55394cc1870766724dd14d
+size 227851840

Pipeline_Head_Fixed.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93251763d5376dad5dc1f78cb0440397abb5a52e346575f1b6b750e958da13eb
-size 827022
+oid sha256:d3ed03e0b14df8f414204eaae42d4aa84137260dd38e887b657eeac3bead9944
+size 702294

Pipeline_Head_Fixed.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be56312ef2dbf57706aab7395fc2d5601ac6fbe6c553fd2b25069eba8da9b3b2
-size 235580992
+oid sha256:58897124dc41ee376bd757521ba368cdcd85249e5f55394cc1870766724dd14d
+size 227851840

Pipeline_Head_Fixed.mlpackage/Manifest.json

@@ -1,18 +1,18 @@
 {
     "fileFormatVersion": "1.0.0",
     "itemInfoEntries": {
-        "7D7B39C7-0AD2-4CD7-B6B0-A7E76DCDE6CA": {
-            "author": "com.apple.CoreML",
-            "description": "CoreML Model Weights",
-            "name": "weights",
-            "path": "com.apple.CoreML/weights"
-        },
-        "B58CA828-CA78-46FC-BD8D-5ABFB5AAEADD": {
+        "BF03DC6C-1762-4CF1-830A-321FCC55F128": {
             "author": "com.apple.CoreML",
             "description": "CoreML Model Specification",
             "name": "model.mlmodel",
             "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "D2F96194-ABA4-45BA-936D-6151E16A1D6B": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
         }
     },
-    "rootModelIdentifier": "B58CA828-CA78-46FC-BD8D-5ABFB5AAEADD"
+    "rootModelIdentifier": "BF03DC6C-1762-4CF1-830A-321FCC55F128"
 }

Pipeline_PreEncoder.mlmodelc/analytics/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f0946b687ccf4274e0228e3bc539e6733c33bf0f4419e28d02220b19a35d884b
+oid sha256:db618db8383557046441bde4fb8e089cffb76db4381908e9c8792be56ad45171
 size 243

Pipeline_PreEncoder.mlmodelc/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:beb4fe86e80615cf79e64e55cb229cc931998b67b51a661bfbdc66204da0ea7b
-size 553
+oid sha256:378cbefc01fa0a778dd9ecec918218a3c842ee87062e8ccb4b56b81abe1ff39b
+size 631

Pipeline_PreEncoder.mlmodelc/metadata.json

@@ -0,0 +1,158 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float32",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 168 × 512)",
+        "shortDescription" : "",
+        "shape" : "[1, 168, 512]",
+        "name" : "pre_encoder_embs",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "pre_encoder_lengths",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 8 × 512)",
+        "shortDescription" : "",
+        "shape" : "[1, 8, 512]",
+        "name" : "chunk_pre_encoder_embs",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "chunk_pre_encoder_lengths",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 7,
+    "mlProgramOperationTypeHistogram" : {
+      "Concat" : 1,
+      "Ios16.cast" : 12,
+      "Ios16.mul" : 12,
+      "Ios16.floorDiv" : 3,
+      "Ios16.less" : 5,
+      "Ios16.sub" : 5,
+      "Transpose" : 1,
+      "Ios16.conv" : 5,
+      "Ios16.add" : 13,
+      "Ios16.relu" : 3,
+      "Tile" : 8,
+      "Ios16.linear" : 1,
+      "Ios16.greaterEqual" : 2,
+      "ExpandDims" : 17,
+      "Ios16.minimum" : 1,
+      "Ios16.maximum" : 1,
+      "Ios16.reshape" : 1,
+      "Ios16.gatherAlongAxis" : 1
+    },
+    "computePrecision" : "Mixed (Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "13.0",
+      "tvOS" : "16.0",
+      "visionOS" : "1.0",
+      "watchOS" : "9.0",
+      "iOS" : "16.0",
+      "macCatalyst" : "16.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-12-25",
+      "com.github.apple.coremltools.source" : "torch==2.9.1",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 64 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 64, 128]",
+        "name" : "chunk",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "chunk_lengths",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 120 × 512)",
+        "shortDescription" : "",
+        "shape" : "[1, 120, 512]",
+        "name" : "spkcache",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "spkcache_lengths",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 40 × 512)",
+        "shortDescription" : "",
+        "shape" : "[1, 40, 512]",
+        "name" : "fifo",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "fifo_lengths",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "Pipeline_PreEncoder",
+    "method" : "predict"
+  }
+]

Pipeline_PreEncoder.mlmodelc/model.mil

@@ -1,7 +1,7 @@
 program(1.0)
-[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3505.3.2"}, {"coremlc-version", "3505.4.1"}, {"coremltools-component-torch", "2.4.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})]
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.9.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
 {
-    func main<ios16>(tensor<fp32, [1, 112, 128]> chunk, tensor<int32, [1]> chunk_lengths, tensor<fp32, [1, 188, 512]> fifo, tensor<int32, [1]> fifo_lengths, tensor<fp32, [1, 188, 512]> spkcache, tensor<int32, [1]> spkcache_lengths) {
+    func main<ios16>(tensor<fp32, [1, 64, 128]> chunk, tensor<int32, [1]> chunk_lengths, tensor<fp32, [1, 40, 512]> fifo, tensor<int32, [1]> fifo_lengths, tensor<fp32, [1, 120, 512]> spkcache, tensor<int32, [1]> spkcache_lengths) {
             tensor<fp32, [256]> model_encoder_pre_encode_conv_0_bias = const()[name = tensor<string, []>("model_encoder_pre_encode_conv_0_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
             tensor<fp32, [256, 1, 3, 3]> model_encoder_pre_encode_conv_0_weight = const()[name = tensor<string, []>("model_encoder_pre_encode_conv_0_weight"), val = tensor<fp32, [256, 1, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1152)))];
             tensor<fp32, [256]> model_encoder_pre_encode_conv_2_bias = const()[name = tensor<string, []>("model_encoder_pre_encode_conv_2_bias"), val = tensor<fp32, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10432)))];
@@ -15,30 +15,30 @@ program(1.0)
             tensor<fp32, [512]> model_encoder_pre_encode_out_bias = const()[name = tensor<string, []>("model_encoder_pre_encode_out_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(557760)))];
             tensor<fp32, [512, 4096]> model_encoder_pre_encode_out_weight = const()[name = tensor<string, []>("model_encoder_pre_encode_out_weight"), val = tensor<fp32, [512, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(559872)))];
             tensor<int32, [1]> tensor_1_axes_0 = const()[name = tensor<string, []>("tensor_1_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp32, [1, 1, 112, 128]> tensor_1 = expand_dims(axes = tensor_1_axes_0, x = chunk)[name = tensor<string, []>("tensor_1")];
-            tensor<string, []> cast_0_dtype_0 = const()[name = tensor<string, []>("cast_0_dtype_0"), val = tensor<string, []>("fp32")];
-            tensor<int32, [1, 112]> expand_dims_0 = const()[name = tensor<string, []>("expand_dims_0"), val = tensor<int32, [1, 112]>([[0, 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]])];
+            tensor<fp32, [1, 1, 64, 128]> tensor_1 = expand_dims(axes = tensor_1_axes_0, x = chunk)[name = tensor<string, []>("tensor_1")];
+            tensor<string, []> current_lengths_1_dtype_0 = const()[name = tensor<string, []>("current_lengths_1_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<int32, [1, 64]> expand_dims_0 = const()[name = tensor<string, []>("expand_dims_0"), val = tensor<int32, [1, 64]>([[0, 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]])];
             tensor<int32, [1]> var_40_axes_0 = const()[name = tensor<string, []>("op_40_axes_0"), val = tensor<int32, [1]>([1])];
             tensor<int32, [1, 1]> var_40 = expand_dims(axes = var_40_axes_0, x = chunk_lengths)[name = tensor<string, []>("op_40")];
-            tensor<bool, [1, 112]> time_mask_1 = less(x = expand_dims_0, y = var_40)[name = tensor<string, []>("time_mask_1")];
+            tensor<bool, [1, 64]> time_mask_1 = less(x = expand_dims_0, y = var_40)[name = tensor<string, []>("time_mask_1")];
             tensor<int32, [1]> var_42_axes_0 = const()[name = tensor<string, []>("op_42_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<bool, [1, 112, 1]> var_42 = expand_dims(axes = var_42_axes_0, x = time_mask_1)[name = tensor<string, []>("op_42")];
+            tensor<bool, [1, 64, 1]> var_42 = expand_dims(axes = var_42_axes_0, x = time_mask_1)[name = tensor<string, []>("op_42")];
             tensor<int32, [3]> var_44_reps_0 = const()[name = tensor<string, []>("op_44_reps_0"), val = tensor<int32, [3]>([1, 1, 128])];
-            tensor<bool, [1, 112, 128]> var_44 = tile(reps = var_44_reps_0, x = var_42)[name = tensor<string, []>("op_44")];
-            tensor<string, []> cast_2_dtype_0 = const()[name = tensor<string, []>("cast_2_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<bool, [1, 64, 128]> var_44 = tile(reps = var_44_reps_0, x = var_42)[name = tensor<string, []>("op_44")];
+            tensor<string, []> mask_1_dtype_0 = const()[name = tensor<string, []>("mask_1_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])];
-            tensor<fp32, [1, 112, 128]> cast_2 = cast(dtype = cast_2_dtype_0, x = var_44)[name = tensor<string, []>("cast_25")];
-            tensor<fp32, [1, 1, 112, 128]> var_50 = expand_dims(axes = var_50_axes_0, x = cast_2)[name = tensor<string, []>("op_50")];
-            tensor<fp32, [1, 1, 112, 128]> input_1 = mul(x = tensor_1, y = var_50)[name = tensor<string, []>("input_1")];
+            tensor<fp32, [1, 64, 128]> mask_1 = cast(dtype = mask_1_dtype_0, x = var_44)[name = tensor<string, []>("cast_11")];
+            tensor<fp32, [1, 1, 64, 128]> var_50 = expand_dims(axes = var_50_axes_0, x = mask_1)[name = tensor<string, []>("op_50")];
+            tensor<fp32, [1, 1, 64, 128]> input_1 = mul(x = tensor_1, y = var_50)[name = tensor<string, []>("input_1")];
             tensor<string, []> tensor_3_pad_type_0 = const()[name = tensor<string, []>("tensor_3_pad_type_0"), val = tensor<string, []>("custom")];
             tensor<int32, [4]> tensor_3_pad_0 = const()[name = tensor<string, []>("tensor_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
             tensor<int32, [2]> tensor_3_strides_0 = const()[name = tensor<string, []>("tensor_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
             tensor<int32, [2]> tensor_3_dilations_0 = const()[name = tensor<string, []>("tensor_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
             tensor<int32, []> tensor_3_groups_0 = const()[name = tensor<string, []>("tensor_3_groups_0"), val = tensor<int32, []>(1)];
-            tensor<fp32, [1, 256, 56, 64]> tensor_3 = conv(bias = model_encoder_pre_encode_conv_0_bias, dilations = tensor_3_dilations_0, groups = tensor_3_groups_0, pad = tensor_3_pad_0, pad_type = tensor_3_pad_type_0, strides = tensor_3_strides_0, weight = model_encoder_pre_encode_conv_0_weight, x = input_1)[name = tensor<string, []>("tensor_3")];
+            tensor<fp32, [1, 256, 32, 64]> tensor_3 = conv(bias = model_encoder_pre_encode_conv_0_bias, dilations = tensor_3_dilations_0, groups = tensor_3_groups_0, pad = tensor_3_pad_0, pad_type = tensor_3_pad_type_0, strides = tensor_3_strides_0, weight = model_encoder_pre_encode_conv_0_weight, x = input_1)[name = tensor<string, []>("tensor_3")];
             tensor<fp32, []> var_61_promoted = const()[name = tensor<string, []>("op_61_promoted"), val = tensor<fp32, []>(0x1p+0)];
-            tensor<fp32, [1]> cast_0 = cast(dtype = cast_0_dtype_0, x = chunk_lengths)[name = tensor<string, []>("cast_26")];
-            tensor<fp32, [1]> var_62 = add(x = cast_0, y = var_61_promoted)[name = tensor<string, []>("op_62")];
+            tensor<fp32, [1]> current_lengths_1 = cast(dtype = current_lengths_1_dtype_0, x = chunk_lengths)[name = tensor<string, []>("cast_12")];
+            tensor<fp32, [1]> var_62 = add(x = current_lengths_1, y = var_61_promoted)[name = tensor<string, []>("op_62")];
             tensor<fp32, []> var_63_promoted = const()[name = tensor<string, []>("op_63_promoted"), val = tensor<fp32, []>(0x1p+0)];
             tensor<fp32, [1]> var_64 = add(x = var_62, y = var_63_promoted)[name = tensor<string, []>("op_64")];
             tensor<fp32, []> var_65_promoted = const()[name = tensor<string, []>("op_65_promoted"), val = tensor<fp32, []>(0x1.8p+1)];
@@ -47,31 +47,31 @@ program(1.0)
             tensor<fp32, [1]> floor_div_0 = floor_div(x = var_66, y = var_21_promoted)[name = tensor<string, []>("floor_div_0")];
             tensor<fp32, []> var_68_promoted = const()[name = tensor<string, []>("op_68_promoted"), val = tensor<fp32, []>(0x1p+0)];
             tensor<fp32, [1]> current_lengths_3 = add(x = floor_div_0, y = var_68_promoted)[name = tensor<string, []>("current_lengths_3")];
-            tensor<string, []> cast_3_dtype_0 = const()[name = tensor<string, []>("cast_3_dtype_0"), val = tensor<string, []>("int32")];
-            tensor<int32, [1, 56]> expand_dims_1 = const()[name = tensor<string, []>("expand_dims_1"), val = tensor<int32, [1, 56]>([[0, 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]])];
+            tensor<string, []> lengths_21_dtype_0 = const()[name = tensor<string, []>("lengths_21_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1, 32]> expand_dims_1 = const()[name = tensor<string, []>("expand_dims_1"), val = tensor<int32, [1, 32]>([[0, 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]])];
             tensor<int32, [1]> var_77_axes_0 = const()[name = tensor<string, []>("op_77_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<int32, [1]> cast_3 = cast(dtype = cast_3_dtype_0, x = current_lengths_3)[name = tensor<string, []>("cast_24")];
-            tensor<int32, [1, 1]> var_77 = expand_dims(axes = var_77_axes_0, x = cast_3)[name = tensor<string, []>("op_77")];
-            tensor<bool, [1, 56]> time_mask_3 = less(x = expand_dims_1, y = var_77)[name = tensor<string, []>("time_mask_3")];
+            tensor<int32, [1]> lengths_21 = cast(dtype = lengths_21_dtype_0, x = current_lengths_3)[name = tensor<string, []>("cast_10")];
+            tensor<int32, [1, 1]> var_77 = expand_dims(axes = var_77_axes_0, x = lengths_21)[name = tensor<string, []>("op_77")];
+            tensor<bool, [1, 32]> time_mask_3 = less(x = expand_dims_1, y = var_77)[name = tensor<string, []>("time_mask_3")];
             tensor<int32, [1]> var_79_axes_0 = const()[name = tensor<string, []>("op_79_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<bool, [1, 56, 1]> var_79 = expand_dims(axes = var_79_axes_0, x = time_mask_3)[name = tensor<string, []>("op_79")];
+            tensor<bool, [1, 32, 1]> var_79 = expand_dims(axes = var_79_axes_0, x = time_mask_3)[name = tensor<string, []>("op_79")];
             tensor<int32, [3]> var_81_reps_0 = const()[name = tensor<string, []>("op_81_reps_0"), val = tensor<int32, [3]>([1, 1, 64])];
-            tensor<bool, [1, 56, 64]> var_81 = tile(reps = var_81_reps_0, x = var_79)[name = tensor<string, []>("op_81")];
-            tensor<string, []> cast_4_dtype_0 = const()[name = tensor<string, []>("cast_4_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<bool, [1, 32, 64]> var_81 = tile(reps = var_81_reps_0, x = var_79)[name = tensor<string, []>("op_81")];
+            tensor<string, []> mask_3_dtype_0 = const()[name = tensor<string, []>("mask_3_dtype_0"), val = tensor<string, []>("fp32")];
             tensor<int32, [1]> var_87_axes_0 = const()[name = tensor<string, []>("op_87_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp32, [1, 56, 64]> cast_4 = cast(dtype = cast_4_dtype_0, x = var_81)[name = tensor<string, []>("cast_23")];
-            tensor<fp32, [1, 1, 56, 64]> var_87 = expand_dims(axes = var_87_axes_0, x = cast_4)[name = tensor<string, []>("op_87")];
+            tensor<fp32, [1, 32, 64]> mask_3 = cast(dtype = mask_3_dtype_0, x = var_81)[name = tensor<string, []>("cast_9")];
+            tensor<fp32, [1, 1, 32, 64]> var_87 = expand_dims(axes = var_87_axes_0, x = mask_3)[name = tensor<string, []>("op_87")];
             tensor<int32, [4]> expanded_mask_3_reps_0 = const()[name = tensor<string, []>("expanded_mask_3_reps_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
-            tensor<fp32, [1, 256, 56, 64]> expanded_mask_3 = tile(reps = expanded_mask_3_reps_0, x = var_87)[name = tensor<string, []>("expanded_mask_3")];
-            tensor<fp32, [1, 256, 56, 64]> input_3 = mul(x = tensor_3, y = expanded_mask_3)[name = tensor<string, []>("input_3")];
-            tensor<fp32, [1, 256, 56, 64]> tensor_5 = relu(x = input_3)[name = tensor<string, []>("tensor_5")];
-            tensor<fp32, [1, 256, 56, 64]> input_5 = mul(x = tensor_5, y = expanded_mask_3)[name = tensor<string, []>("input_5")];
+            tensor<fp32, [1, 256, 32, 64]> expanded_mask_3 = tile(reps = expanded_mask_3_reps_0, x = var_87)[name = tensor<string, []>("expanded_mask_3")];
+            tensor<fp32, [1, 256, 32, 64]> input_3 = mul(x = tensor_3, y = expanded_mask_3)[name = tensor<string, []>("input_3")];
+            tensor<fp32, [1, 256, 32, 64]> tensor_5 = relu(x = input_3)[name = tensor<string, []>("tensor_5")];
+            tensor<fp32, [1, 256, 32, 64]> input_5 = mul(x = tensor_5, y = expanded_mask_3)[name = tensor<string, []>("input_5")];
             tensor<string, []> tensor_7_pad_type_0 = const()[name = tensor<string, []>("tensor_7_pad_type_0"), val = tensor<string, []>("custom")];
             tensor<int32, [4]> tensor_7_pad_0 = const()[name = tensor<string, []>("tensor_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
             tensor<int32, [2]> tensor_7_strides_0 = const()[name = tensor<string, []>("tensor_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
             tensor<int32, []> tensor_7_groups_0 = const()[name = tensor<string, []>("tensor_7_groups_0"), val = tensor<int32, []>(256)];
             tensor<int32, [2]> tensor_7_dilations_0 = const()[name = tensor<string, []>("tensor_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
-            tensor<fp32, [1, 256, 28, 32]> tensor_7 = conv(bias = model_encoder_pre_encode_conv_2_bias, dilations = tensor_7_dilations_0, groups = tensor_7_groups_0, pad = tensor_7_pad_0, pad_type = tensor_7_pad_type_0, strides = tensor_7_strides_0, weight = model_encoder_pre_encode_conv_2_weight, x = input_5)[name = tensor<string, []>("tensor_7")];
+            tensor<fp32, [1, 256, 16, 32]> tensor_7 = conv(bias = model_encoder_pre_encode_conv_2_bias, dilations = tensor_7_dilations_0, groups = tensor_7_groups_0, pad = tensor_7_pad_0, pad_type = tensor_7_pad_type_0, strides = tensor_7_strides_0, weight = model_encoder_pre_encode_conv_2_weight, x = input_5)[name = tensor<string, []>("tensor_7")];
             tensor<fp32, []> var_107_promoted = const()[name = tensor<string, []>("op_107_promoted"), val = tensor<fp32, []>(0x1p+0)];
             tensor<fp32, [1]> var_108 = add(x = current_lengths_3, y = var_107_promoted)[name = tensor<string, []>("op_108")];
             tensor<fp32, []> var_109_promoted = const()[name = tensor<string, []>("op_109_promoted"), val = tensor<fp32, []>(0x1p+0)];
@@ -82,38 +82,38 @@ program(1.0)
             tensor<fp32, [1]> floor_div_1 = floor_div(x = var_112, y = var_21_promoted_1)[name = tensor<string, []>("floor_div_1")];
             tensor<fp32, []> var_114_promoted = const()[name = tensor<string, []>("op_114_promoted"), val = tensor<fp32, []>(0x1p+0)];
             tensor<fp32, [1]> current_lengths_5 = add(x = floor_div_1, y = var_114_promoted)[name = tensor<string, []>("current_lengths_5")];
-            tensor<string, []> cast_5_dtype_0 = const()[name = tensor<string, []>("cast_5_dtype_0"), val = tensor<string, []>("int32")];
-            tensor<int32, [1, 28]> expand_dims_2 = const()[name = tensor<string, []>("expand_dims_2"), val = tensor<int32, [1, 28]>([[0, 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]])];
+            tensor<string, []> lengths_23_dtype_0 = const()[name = tensor<string, []>("lengths_23_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1, 16]> expand_dims_2 = const()[name = tensor<string, []>("expand_dims_2"), val = tensor<int32, [1, 16]>([[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]])];
             tensor<int32, [1]> var_123_axes_0 = const()[name = tensor<string, []>("op_123_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<int32, [1]> cast_5 = cast(dtype = cast_5_dtype_0, x = current_lengths_5)[name = tensor<string, []>("cast_22")];
-            tensor<int32, [1, 1]> var_123 = expand_dims(axes = var_123_axes_0, x = cast_5)[name = tensor<string, []>("op_123")];
-            tensor<bool, [1, 28]> time_mask_5 = less(x = expand_dims_2, y = var_123)[name = tensor<string, []>("time_mask_5")];
+            tensor<int32, [1]> lengths_23 = cast(dtype = lengths_23_dtype_0, x = current_lengths_5)[name = tensor<string, []>("cast_8")];
+            tensor<int32, [1, 1]> var_123 = expand_dims(axes = var_123_axes_0, x = lengths_23)[name = tensor<string, []>("op_123")];
+            tensor<bool, [1, 16]> time_mask_5 = less(x = expand_dims_2, y = var_123)[name = tensor<string, []>("time_mask_5")];
             tensor<int32, [1]> var_125_axes_0 = const()[name = tensor<string, []>("op_125_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<bool, [1, 28, 1]> var_125 = expand_dims(axes = var_125_axes_0, x = time_mask_5)[name = tensor<string, []>("op_125")];
+            tensor<bool, [1, 16, 1]> var_125 = expand_dims(axes = var_125_axes_0, x = time_mask_5)[name = tensor<string, []>("op_125")];
             tensor<int32, [3]> var_127_reps_0 = const()[name = tensor<string, []>("op_127_reps_0"), val = tensor<int32, [3]>([1, 1, 32])];
-            tensor<bool, [1, 28, 32]> var_127 = tile(reps = var_127_reps_0, x = var_125)[name = tensor<string, []>("op_127")];
-            tensor<string, []> cast_6_dtype_0 = const()[name = tensor<string, []>("cast_6_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<bool, [1, 16, 32]> var_127 = tile(reps = var_127_reps_0, x = var_125)[name = tensor<string, []>("op_127")];
+            tensor<string, []> mask_5_dtype_0 = const()[name = tensor<string, []>("mask_5_dtype_0"), val = tensor<string, []>("fp32")];
             tensor<int32, [1]> var_133_axes_0 = const()[name = tensor<string, []>("op_133_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp32, [1, 28, 32]> cast_6 = cast(dtype = cast_6_dtype_0, x = var_127)[name = tensor<string, []>("cast_21")];
-            tensor<fp32, [1, 1, 28, 32]> var_133 = expand_dims(axes = var_133_axes_0, x = cast_6)[name = tensor<string, []>("op_133")];
+            tensor<fp32, [1, 16, 32]> mask_5 = cast(dtype = mask_5_dtype_0, x = var_127)[name = tensor<string, []>("cast_7")];
+            tensor<fp32, [1, 1, 16, 32]> var_133 = expand_dims(axes = var_133_axes_0, x = mask_5)[name = tensor<string, []>("op_133")];
             tensor<int32, [4]> expanded_mask_7_reps_0 = const()[name = tensor<string, []>("expanded_mask_7_reps_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
-            tensor<fp32, [1, 256, 28, 32]> expanded_mask_7 = tile(reps = expanded_mask_7_reps_0, x = var_133)[name = tensor<string, []>("expanded_mask_7")];
-            tensor<fp32, [1, 256, 28, 32]> input_7 = mul(x = tensor_7, y = expanded_mask_7)[name = tensor<string, []>("input_7")];
+            tensor<fp32, [1, 256, 16, 32]> expanded_mask_7 = tile(reps = expanded_mask_7_reps_0, x = var_133)[name = tensor<string, []>("expanded_mask_7")];
+            tensor<fp32, [1, 256, 16, 32]> input_7 = mul(x = tensor_7, y = expanded_mask_7)[name = tensor<string, []>("input_7")];
             tensor<string, []> tensor_9_pad_type_0 = const()[name = tensor<string, []>("tensor_9_pad_type_0"), val = tensor<string, []>("valid")];
             tensor<int32, [2]> tensor_9_strides_0 = const()[name = tensor<string, []>("tensor_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
             tensor<int32, [4]> tensor_9_pad_0 = const()[name = tensor<string, []>("tensor_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
             tensor<int32, [2]> tensor_9_dilations_0 = const()[name = tensor<string, []>("tensor_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
             tensor<int32, []> tensor_9_groups_0 = const()[name = tensor<string, []>("tensor_9_groups_0"), val = tensor<int32, []>(1)];
-            tensor<fp32, [1, 256, 28, 32]> tensor_9 = conv(bias = model_encoder_pre_encode_conv_3_bias, dilations = tensor_9_dilations_0, groups = tensor_9_groups_0, pad = tensor_9_pad_0, pad_type = tensor_9_pad_type_0, strides = tensor_9_strides_0, weight = model_encoder_pre_encode_conv_3_weight, x = input_7)[name = tensor<string, []>("tensor_9")];
-            tensor<fp32, [1, 256, 28, 32]> input_9 = mul(x = tensor_9, y = expanded_mask_7)[name = tensor<string, []>("input_9")];
-            tensor<fp32, [1, 256, 28, 32]> tensor_11 = relu(x = input_9)[name = tensor<string, []>("tensor_11")];
-            tensor<fp32, [1, 256, 28, 32]> input_11 = mul(x = tensor_11, y = expanded_mask_7)[name = tensor<string, []>("input_11")];
+            tensor<fp32, [1, 256, 16, 32]> tensor_9 = conv(bias = model_encoder_pre_encode_conv_3_bias, dilations = tensor_9_dilations_0, groups = tensor_9_groups_0, pad = tensor_9_pad_0, pad_type = tensor_9_pad_type_0, strides = tensor_9_strides_0, weight = model_encoder_pre_encode_conv_3_weight, x = input_7)[name = tensor<string, []>("tensor_9")];
+            tensor<fp32, [1, 256, 16, 32]> input_9 = mul(x = tensor_9, y = expanded_mask_7)[name = tensor<string, []>("input_9")];
+            tensor<fp32, [1, 256, 16, 32]> tensor_11 = relu(x = input_9)[name = tensor<string, []>("tensor_11")];
+            tensor<fp32, [1, 256, 16, 32]> input_11 = mul(x = tensor_11, y = expanded_mask_7)[name = tensor<string, []>("input_11")];
             tensor<string, []> tensor_13_pad_type_0 = const()[name = tensor<string, []>("tensor_13_pad_type_0"), val = tensor<string, []>("custom")];
             tensor<int32, [4]> tensor_13_pad_0 = const()[name = tensor<string, []>("tensor_13_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
             tensor<int32, [2]> tensor_13_strides_0 = const()[name = tensor<string, []>("tensor_13_strides_0"), val = tensor<int32, [2]>([2, 2])];
             tensor<int32, []> tensor_13_groups_0 = const()[name = tensor<string, []>("tensor_13_groups_0"), val = tensor<int32, []>(256)];
             tensor<int32, [2]> tensor_13_dilations_0 = const()[name = tensor<string, []>("tensor_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
-            tensor<fp32, [1, 256, 14, 16]> tensor_13 = conv(bias = model_encoder_pre_encode_conv_5_bias, dilations = tensor_13_dilations_0, groups = tensor_13_groups_0, pad = tensor_13_pad_0, pad_type = tensor_13_pad_type_0, strides = tensor_13_strides_0, weight = model_encoder_pre_encode_conv_5_weight, x = input_11)[name = tensor<string, []>("tensor_13")];
+            tensor<fp32, [1, 256, 8, 16]> tensor_13 = conv(bias = model_encoder_pre_encode_conv_5_bias, dilations = tensor_13_dilations_0, groups = tensor_13_groups_0, pad = tensor_13_pad_0, pad_type = tensor_13_pad_type_0, strides = tensor_13_strides_0, weight = model_encoder_pre_encode_conv_5_weight, x = input_11)[name = tensor<string, []>("tensor_13")];
             tensor<fp32, []> var_168_promoted = const()[name = tensor<string, []>("op_168_promoted"), val = tensor<fp32, []>(0x1p+0)];
             tensor<fp32, [1]> var_169 = add(x = current_lengths_5, y = var_168_promoted)[name = tensor<string, []>("op_169")];
             tensor<fp32, []> var_170_promoted = const()[name = tensor<string, []>("op_170_promoted"), val = tensor<fp32, []>(0x1p+0)];
@@ -124,78 +124,78 @@ program(1.0)
             tensor<fp32, [1]> floor_div_2 = floor_div(x = var_173, y = var_21_promoted_2)[name = tensor<string, []>("floor_div_2")];
             tensor<fp32, []> var_175_promoted = const()[name = tensor<string, []>("op_175_promoted"), val = tensor<fp32, []>(0x1p+0)];
             tensor<fp32, [1]> current_lengths = add(x = floor_div_2, y = var_175_promoted)[name = tensor<string, []>("current_lengths")];
-            tensor<string, []> cast_7_dtype_0 = const()[name = tensor<string, []>("cast_7_dtype_0"), val = tensor<string, []>("int32")];
-            tensor<int32, [1, 14]> expand_dims_3 = const()[name = tensor<string, []>("expand_dims_3"), val = tensor<int32, [1, 14]>([[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13]])];
+            tensor<string, []> lengths_dtype_0 = const()[name = tensor<string, []>("lengths_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1, 8]> expand_dims_3 = const()[name = tensor<string, []>("expand_dims_3"), val = tensor<int32, [1, 8]>([[0, 1, 2, 3, 4, 5, 6, 7]])];
             tensor<int32, [1]> var_184_axes_0 = const()[name = tensor<string, []>("op_184_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<int32, [1]> cast_7 = cast(dtype = cast_7_dtype_0, x = current_lengths)[name = tensor<string, []>("cast_20")];
-            tensor<int32, [1, 1]> var_184 = expand_dims(axes = var_184_axes_0, x = cast_7)[name = tensor<string, []>("op_184")];
-            tensor<bool, [1, 14]> time_mask = less(x = expand_dims_3, y = var_184)[name = tensor<string, []>("time_mask")];
+            tensor<int32, [1]> lengths = cast(dtype = lengths_dtype_0, x = current_lengths)[name = tensor<string, []>("cast_6")];
+            tensor<int32, [1, 1]> var_184 = expand_dims(axes = var_184_axes_0, x = lengths)[name = tensor<string, []>("op_184")];
+            tensor<bool, [1, 8]> time_mask = less(x = expand_dims_3, y = var_184)[name = tensor<string, []>("time_mask")];
             tensor<int32, [1]> var_186_axes_0 = const()[name = tensor<string, []>("op_186_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<bool, [1, 14, 1]> var_186 = expand_dims(axes = var_186_axes_0, x = time_mask)[name = tensor<string, []>("op_186")];
+            tensor<bool, [1, 8, 1]> var_186 = expand_dims(axes = var_186_axes_0, x = time_mask)[name = tensor<string, []>("op_186")];
             tensor<int32, [3]> var_188_reps_0 = const()[name = tensor<string, []>("op_188_reps_0"), val = tensor<int32, [3]>([1, 1, 16])];
-            tensor<bool, [1, 14, 16]> var_188 = tile(reps = var_188_reps_0, x = var_186)[name = tensor<string, []>("op_188")];
-            tensor<string, []> cast_8_dtype_0 = const()[name = tensor<string, []>("cast_8_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<bool, [1, 8, 16]> var_188 = tile(reps = var_188_reps_0, x = var_186)[name = tensor<string, []>("op_188")];
+            tensor<string, []> mask_dtype_0 = const()[name = tensor<string, []>("mask_dtype_0"), val = tensor<string, []>("fp32")];
             tensor<int32, [1]> var_194_axes_0 = const()[name = tensor<string, []>("op_194_axes_0"), val = tensor<int32, [1]>([1])];
-            tensor<fp32, [1, 14, 16]> cast_8 = cast(dtype = cast_8_dtype_0, x = var_188)[name = tensor<string, []>("cast_19")];
-            tensor<fp32, [1, 1, 14, 16]> var_194 = expand_dims(axes = var_194_axes_0, x = cast_8)[name = tensor<string, []>("op_194")];
+            tensor<fp32, [1, 8, 16]> mask = cast(dtype = mask_dtype_0, x = var_188)[name = tensor<string, []>("cast_5")];
+            tensor<fp32, [1, 1, 8, 16]> var_194 = expand_dims(axes = var_194_axes_0, x = mask)[name = tensor<string, []>("op_194")];
             tensor<int32, [4]> expanded_mask_13_reps_0 = const()[name = tensor<string, []>("expanded_mask_13_reps_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
-            tensor<fp32, [1, 256, 14, 16]> expanded_mask_13 = tile(reps = expanded_mask_13_reps_0, x = var_194)[name = tensor<string, []>("expanded_mask_13")];
-            tensor<fp32, [1, 256, 14, 16]> input_13 = mul(x = tensor_13, y = expanded_mask_13)[name = tensor<string, []>("input_13")];
+            tensor<fp32, [1, 256, 8, 16]> expanded_mask_13 = tile(reps = expanded_mask_13_reps_0, x = var_194)[name = tensor<string, []>("expanded_mask_13")];
+            tensor<fp32, [1, 256, 8, 16]> input_13 = mul(x = tensor_13, y = expanded_mask_13)[name = tensor<string, []>("input_13")];
             tensor<string, []> tensor_15_pad_type_0 = const()[name = tensor<string, []>("tensor_15_pad_type_0"), val = tensor<string, []>("valid")];
             tensor<int32, [2]> tensor_15_strides_0 = const()[name = tensor<string, []>("tensor_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
             tensor<int32, [4]> tensor_15_pad_0 = const()[name = tensor<string, []>("tensor_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
             tensor<int32, [2]> tensor_15_dilations_0 = const()[name = tensor<string, []>("tensor_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
             tensor<int32, []> tensor_15_groups_0 = const()[name = tensor<string, []>("tensor_15_groups_0"), val = tensor<int32, []>(1)];
-            tensor<fp32, [1, 256, 14, 16]> tensor_15 = conv(bias = model_encoder_pre_encode_conv_6_bias, dilations = tensor_15_dilations_0, groups = tensor_15_groups_0, pad = tensor_15_pad_0, pad_type = tensor_15_pad_type_0, strides = tensor_15_strides_0, weight = model_encoder_pre_encode_conv_6_weight, x = input_13)[name = tensor<string, []>("tensor_15")];
-            tensor<fp32, [1, 256, 14, 16]> input_15 = mul(x = tensor_15, y = expanded_mask_13)[name = tensor<string, []>("input_15")];
-            tensor<fp32, [1, 256, 14, 16]> tensor_workaround = relu(x = input_15)[name = tensor<string, []>("tensor_workaround")];
-            tensor<fp32, [1, 256, 14, 16]> x = mul(x = tensor_workaround, y = expanded_mask_13)[name = tensor<string, []>("x")];
+            tensor<fp32, [1, 256, 8, 16]> tensor_15 = conv(bias = model_encoder_pre_encode_conv_6_bias, dilations = tensor_15_dilations_0, groups = tensor_15_groups_0, pad = tensor_15_pad_0, pad_type = tensor_15_pad_type_0, strides = tensor_15_strides_0, weight = model_encoder_pre_encode_conv_6_weight, x = input_13)[name = tensor<string, []>("tensor_15")];
+            tensor<fp32, [1, 256, 8, 16]> input_15 = mul(x = tensor_15, y = expanded_mask_13)[name = tensor<string, []>("input_15")];
+            tensor<fp32, [1, 256, 8, 16]> tensor_workaround = relu(x = input_15)[name = tensor<string, []>("tensor_workaround")];
+            tensor<fp32, [1, 256, 8, 16]> x = mul(x = tensor_workaround, y = expanded_mask_13)[name = tensor<string, []>("x")];
             tensor<int32, [4]> var_228_perm_0 = const()[name = tensor<string, []>("op_228_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
-            tensor<int32, [3]> var_229 = const()[name = tensor<string, []>("op_229"), val = tensor<int32, [3]>([1, 14, -1])];
-            tensor<fp32, [1, 14, 256, 16]> var_228 = transpose(perm = var_228_perm_0, x = x)[name = tensor<string, []>("transpose_0")];
-            tensor<fp32, [1, 14, 4096]> input = reshape(shape = var_229, x = var_228)[name = tensor<string, []>("input")];
-            tensor<fp32, [1, 14, 512]> chunk_embs_in = linear(bias = model_encoder_pre_encode_out_bias, weight = model_encoder_pre_encode_out_weight, x = input)[name = tensor<string, []>("linear_0")];
-            tensor<string, []> cast_11_dtype_0 = const()[name = tensor<string, []>("cast_11_dtype_0"), val = tensor<string, []>("int32")];
-            tensor<int32, [1]> size0 = const()[name = tensor<string, []>("size0"), val = tensor<int32, [1]>([188])];
-            tensor<int32, [1]> size1 = const()[name = tensor<string, []>("size1"), val = tensor<int32, [1]>([188])];
+            tensor<int32, [3]> var_229 = const()[name = tensor<string, []>("op_229"), val = tensor<int32, [3]>([1, 8, -1])];
+            tensor<fp32, [1, 8, 256, 16]> var_228 = transpose(perm = var_228_perm_0, x = x)[name = tensor<string, []>("transpose_0")];
+            tensor<fp32, [1, 8, 4096]> input = reshape(shape = var_229, x = var_228)[name = tensor<string, []>("input")];
+            tensor<fp32, [1, 8, 512]> chunk_pre_encoder_embs = linear(bias = model_encoder_pre_encode_out_bias, weight = model_encoder_pre_encode_out_weight, x = input)[name = tensor<string, []>("linear_0")];
+            tensor<string, []> var_241_dtype_0 = const()[name = tensor<string, []>("op_241_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1]> size0 = const()[name = tensor<string, []>("size0"), val = tensor<int32, [1]>([120])];
+            tensor<int32, [1]> size1 = const()[name = tensor<string, []>("size1"), val = tensor<int32, [1]>([40])];
             tensor<int32, []> var_264 = const()[name = tensor<string, []>("op_264"), val = tensor<int32, []>(1)];
             tensor<bool, []> full_concat_interleave_0 = const()[name = tensor<string, []>("full_concat_interleave_0"), val = tensor<bool, []>(false)];
-            tensor<fp32, [1, 390, 512]> full_concat = concat(axis = var_264, interleave = full_concat_interleave_0, values = (spkcache, fifo, chunk_embs_in))[name = tensor<string, []>("full_concat")];
+            tensor<fp32, [1, 168, 512]> full_concat = concat(axis = var_264, interleave = full_concat_interleave_0, values = (spkcache, fifo, chunk_pre_encoder_embs))[name = tensor<string, []>("full_concat")];
             tensor<int32, [1]> var_273 = add(x = spkcache_lengths, y = fifo_lengths)[name = tensor<string, []>("op_273")];
-            tensor<int32, [1]> chunk_lens_in = cast(dtype = cast_11_dtype_0, x = current_lengths)[name = tensor<string, []>("cast_18")];
-            tensor<int32, [1]> pre_encoder_lengths = add(x = var_273, y = chunk_lens_in)[name = tensor<string, []>("total_length")];
-            tensor<int32, [390]> out_pos = const()[name = tensor<string, []>("out_pos"), val = tensor<int32, [390]>([0, 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])];
-            tensor<bool, [390]> var_284 = greater_equal(x = out_pos, y = spkcache_lengths)[name = tensor<string, []>("op_284")];
-            tensor<string, []> cast_12_dtype_0 = const()[name = tensor<string, []>("cast_12_dtype_0"), val = tensor<string, []>("int32")];
-            tensor<bool, [390]> var_290 = greater_equal(x = out_pos, y = var_273)[name = tensor<string, []>("op_290")];
-            tensor<string, []> cast_13_dtype_0 = const()[name = tensor<string, []>("cast_13_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1]> chunk_pre_encoder_lengths = cast(dtype = var_241_dtype_0, x = current_lengths)[name = tensor<string, []>("cast_4")];
+            tensor<int32, [1]> pre_encoder_lengths = add(x = var_273, y = chunk_pre_encoder_lengths)[name = tensor<string, []>("total_length")];
+            tensor<int32, [168]> out_pos = const()[name = tensor<string, []>("out_pos"), val = tensor<int32, [168]>([0, 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])];
+            tensor<bool, [168]> var_284 = greater_equal(x = out_pos, y = spkcache_lengths)[name = tensor<string, []>("op_284")];
+            tensor<string, []> in_seg1_or_2_dtype_0 = const()[name = tensor<string, []>("in_seg1_or_2_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<bool, [168]> var_290 = greater_equal(x = out_pos, y = var_273)[name = tensor<string, []>("op_290")];
+            tensor<string, []> in_seg2_dtype_0 = const()[name = tensor<string, []>("in_seg2_dtype_0"), val = tensor<string, []>("int32")];
             tensor<int32, [1]> var_297 = sub(x = size0, y = spkcache_lengths)[name = tensor<string, []>("op_297")];
-            tensor<int32, [390]> cast_12 = cast(dtype = cast_12_dtype_0, x = var_284)[name = tensor<string, []>("cast_17")];
-            tensor<int32, [390]> var_298 = mul(x = cast_12, y = var_297)[name = tensor<string, []>("op_298")];
+            tensor<int32, [168]> in_seg1_or_2 = cast(dtype = in_seg1_or_2_dtype_0, x = var_284)[name = tensor<string, []>("cast_3")];
+            tensor<int32, [168]> var_298 = mul(x = in_seg1_or_2, y = var_297)[name = tensor<string, []>("op_298")];
             tensor<int32, [1]> var_300 = sub(x = size1, y = fifo_lengths)[name = tensor<string, []>("op_300")];
-            tensor<int32, [390]> cast_13 = cast(dtype = cast_13_dtype_0, x = var_290)[name = tensor<string, []>("cast_16")];
-            tensor<int32, [390]> var_301 = mul(x = cast_13, y = var_300)[name = tensor<string, []>("op_301")];
-            tensor<int32, [390]> offset = add(x = var_298, y = var_301)[name = tensor<string, []>("offset")];
-            tensor<int32, [390]> var_305 = add(x = out_pos, y = offset)[name = tensor<string, []>("op_305")];
-            tensor<int32, []> var_309 = const()[name = tensor<string, []>("op_309"), val = tensor<int32, []>(389)];
+            tensor<int32, [168]> in_seg2 = cast(dtype = in_seg2_dtype_0, x = var_290)[name = tensor<string, []>("cast_2")];
+            tensor<int32, [168]> var_301 = mul(x = in_seg2, y = var_300)[name = tensor<string, []>("op_301")];
+            tensor<int32, [168]> offset = add(x = var_298, y = var_301)[name = tensor<string, []>("offset")];
+            tensor<int32, [168]> var_305 = add(x = out_pos, y = offset)[name = tensor<string, []>("op_305")];
+            tensor<int32, []> var_309 = const()[name = tensor<string, []>("op_309"), val = tensor<int32, []>(167)];
             tensor<int32, []> var_310 = const()[name = tensor<string, []>("op_310"), val = tensor<int32, []>(0)];
-            tensor<int32, [390]> minimum_0 = minimum(x = var_305, y = var_309)[name = tensor<string, []>("minimum_0")];
-            tensor<int32, [390]> maximum_0 = maximum(x = minimum_0, y = var_310)[name = tensor<string, []>("maximum_0")];
+            tensor<int32, [168]> minimum_0 = minimum(x = var_305, y = var_309)[name = tensor<string, []>("minimum_0")];
+            tensor<int32, [168]> maximum_0 = maximum(x = minimum_0, y = var_310)[name = tensor<string, []>("maximum_0")];
             tensor<int32, [1]> var_313_axes_0 = const()[name = tensor<string, []>("op_313_axes_0"), val = tensor<int32, [1]>([0])];
-            tensor<int32, [1, 390]> var_313 = expand_dims(axes = var_313_axes_0, x = maximum_0)[name = tensor<string, []>("op_313")];
+            tensor<int32, [1, 168]> var_313 = expand_dims(axes = var_313_axes_0, x = maximum_0)[name = tensor<string, []>("op_313")];
             tensor<int32, [1]> var_315_axes_0 = const()[name = tensor<string, []>("op_315_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<int32, [1, 390, 1]> var_315 = expand_dims(axes = var_315_axes_0, x = var_313)[name = tensor<string, []>("op_315")];
+            tensor<int32, [1, 168, 1]> var_315 = expand_dims(axes = var_315_axes_0, x = var_313)[name = tensor<string, []>("op_315")];
             tensor<int32, [3]> gather_idx_reps_0 = const()[name = tensor<string, []>("gather_idx_reps_0"), val = tensor<int32, [3]>([1, 1, 512])];
-            tensor<int32, [1, 390, 512]> gather_idx = tile(reps = gather_idx_reps_0, x = var_315)[name = tensor<string, []>("gather_idx")];
+            tensor<int32, [1, 168, 512]> gather_idx = tile(reps = gather_idx_reps_0, x = var_315)[name = tensor<string, []>("gather_idx")];
             tensor<int32, []> var_320 = const()[name = tensor<string, []>("op_320"), val = tensor<int32, []>(1)];
-            tensor<fp32, [1, 390, 512]> output = gather_along_axis(axis = var_320, indices = gather_idx, x = full_concat)[name = tensor<string, []>("output")];
-            tensor<bool, [390]> var_323 = less(x = out_pos, y = pre_encoder_lengths)[name = tensor<string, []>("op_323")];
-            tensor<string, []> cast_14_dtype_0 = const()[name = tensor<string, []>("cast_14_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, [1, 168, 512]> output = gather_along_axis(axis = var_320, indices = gather_idx, x = full_concat)[name = tensor<string, []>("output")];
+            tensor<bool, [168]> var_323 = less(x = out_pos, y = pre_encoder_lengths)[name = tensor<string, []>("op_323")];
+            tensor<string, []> var_328_dtype_0 = const()[name = tensor<string, []>("op_328_dtype_0"), val = tensor<string, []>("fp32")];
             tensor<int32, [1]> var_330_axes_0 = const()[name = tensor<string, []>("op_330_axes_0"), val = tensor<int32, [1]>([0])];
-            tensor<fp32, [390]> cast_14 = cast(dtype = cast_14_dtype_0, x = var_323)[name = tensor<string, []>("cast_15")];
-            tensor<fp32, [1, 390]> var_330 = expand_dims(axes = var_330_axes_0, x = cast_14)[name = tensor<string, []>("op_330")];
+            tensor<fp32, [168]> var_328 = cast(dtype = var_328_dtype_0, x = var_323)[name = tensor<string, []>("cast_1")];
+            tensor<fp32, [1, 168]> var_330 = expand_dims(axes = var_330_axes_0, x = var_328)[name = tensor<string, []>("op_330")];
             tensor<int32, [1]> var_332_axes_0 = const()[name = tensor<string, []>("op_332_axes_0"), val = tensor<int32, [1]>([-1])];
-            tensor<fp32, [1, 390, 1]> var_332 = expand_dims(axes = var_332_axes_0, x = var_330)[name = tensor<string, []>("op_332")];
-            tensor<fp32, [1, 390, 512]> pre_encoder_embs = mul(x = output, y = var_332)[name = tensor<string, []>("op_333")];
-        } -> (pre_encoder_embs, pre_encoder_lengths, chunk_embs_in, chunk_lens_in);
+            tensor<fp32, [1, 168, 1]> var_332 = expand_dims(axes = var_332_axes_0, x = var_330)[name = tensor<string, []>("op_332")];
+            tensor<fp32, [1, 168, 512]> pre_encoder_embs = mul(x = output, y = var_332)[name = tensor<string, []>("op_333")];
+        } -> (pre_encoder_embs, pre_encoder_lengths, chunk_pre_encoder_embs, chunk_pre_encoder_lengths);
 }

Pipeline_PreEncoder.mlmodelc/weights/weight.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:88a98803e35186b1dfb41d7f748f7cee5093bb6efeb117f56953c17549792fa4
+oid sha256:ad40d62ccd7a0943d2cd9cc8eeee7f27116e58cf6532ab43196b34142fc86583
 size 8948544

Pipeline_PreEncoder.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb3f36b3b9d3f63e7a4f89a8848c6e3bc1a4a983786a832ea2c60cc395525cc2
-size 26802
+oid sha256:283ba65d5f9ef05a017eeb3a1713e8acf4fb176abdc5155851442be91015514b
+size 26507

Pipeline_PreEncoder.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:88a98803e35186b1dfb41d7f748f7cee5093bb6efeb117f56953c17549792fa4
+oid sha256:ad40d62ccd7a0943d2cd9cc8eeee7f27116e58cf6532ab43196b34142fc86583
 size 8948544

Pipeline_PreEncoder.mlpackage/Manifest.json

@@ -1,18 +1,18 @@
 {
     "fileFormatVersion": "1.0.0",
     "itemInfoEntries": {
-        "6894C507-E04A-4096-A90F-9AB0F58870E0": {
-            "author": "com.apple.CoreML",
-            "description": "CoreML Model Specification",
-            "name": "model.mlmodel",
-            "path": "com.apple.CoreML/model.mlmodel"
-        },
-        "E6A26F48-2E38-4E3A-AF2B-89704AAA4B4C": {
+        "607771B0-3143-4C73-95A8-09D62FBBD09E": {
             "author": "com.apple.CoreML",
             "description": "CoreML Model Weights",
             "name": "weights",
             "path": "com.apple.CoreML/weights"
+        },
+        "7D2D1235-8128-4E73-9A25-C17071167C4F": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
         }
     },
-    "rootModelIdentifier": "6894C507-E04A-4096-A90F-9AB0F58870E0"
+    "rootModelIdentifier": "7D2D1235-8128-4E73-9A25-C17071167C4F"
 }

Pipeline_Preprocessor.mlmodelc/analytics/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eee5506c26dd1453734200ef08e1a263599e25fbdd433ecc425e7d8fd3c39641
+oid sha256:963f42f48df0d5165810bc49fb19ac56e145c7029a45db2cbd258ca4cb32853d
 size 243

Pipeline_Preprocessor.mlmodelc/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15271d7ebb0d1f8f82ae60468e6a554b71d386f6f5f717b00332b9ace990be16
-size 374
+oid sha256:1071f8d3419e2d2a91d59dfb3d956c24234f6e507349f76ceb7f197754b3b806
+size 431

Pipeline_Preprocessor.mlmodelc/metadata.json

@@ -0,0 +1,101 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float32",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 128 × 80)",
+        "shortDescription" : "",
+        "shape" : "[1, 128, 80]",
+        "name" : "features",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "feature_lengths",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 7,
+    "mlProgramOperationTypeHistogram" : {
+      "Ios16.floorDiv" : 1,
+      "Identity" : 1,
+      "Select" : 3,
+      "Concat" : 1,
+      "Ios16.add" : 2,
+      "Ios16.greaterEqual" : 1,
+      "Ios16.reduceSum" : 1,
+      "Pad" : 2,
+      "Ios16.log" : 1,
+      "ExpandDims" : 5,
+      "Ios16.sub" : 2,
+      "Ios16.less" : 1,
+      "Ios16.conv" : 2,
+      "Ios16.matmul" : 1,
+      "Ios16.pow" : 1,
+      "SliceByIndex" : 3,
+      "Stack" : 1,
+      "Ios16.reshape" : 2,
+      "Ios16.equal" : 1,
+      "Ios16.mul" : 1,
+      "Ios16.logicalNot" : 1
+    },
+    "computePrecision" : "Mixed (Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "13.0",
+      "tvOS" : "16.0",
+      "visionOS" : "1.0",
+      "watchOS" : "9.0",
+      "iOS" : "16.0",
+      "macCatalyst" : "16.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.conversion_date" : "2025-12-25",
+      "com.github.apple.coremltools.source" : "torch==2.9.1",
+      "com.github.apple.coremltools.version" : "9.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float32",
+        "formattedType" : "MultiArray (Float32 1 × 10480)",
+        "shortDescription" : "",
+        "shape" : "[1, 10480]",
+        "name" : "audio_signal",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Int32",
+        "formattedType" : "MultiArray (Int32 1)",
+        "shortDescription" : "",
+        "shape" : "[1]",
+        "name" : "length",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "Pipeline_Preprocessor",
+    "method" : "predict"
+  }
+]

Pipeline_Preprocessor.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ba45e5189ff1a12d01f2ebc6fc5db8ab7ad63cedcbb9847515ad0e7881daa0e
-size 48673
+oid sha256:9b5d1719e5213fc75bf109c36af9eda2e043baa0ddafef44855b746b6ccfb585
+size 32243

Pipeline_Preprocessor.mlpackage/Manifest.json

@@ -1,18 +1,18 @@
 {
     "fileFormatVersion": "1.0.0",
     "itemInfoEntries": {
-        "43321C07-C241-4F12-89F5-DF8385087F7C": {
-            "author": "com.apple.CoreML",
-            "description": "CoreML Model Weights",
-            "name": "weights",
-            "path": "com.apple.CoreML/weights"
-        },
-        "505C3394-5B98-45F7-8F05-9C513B04AFCB": {
+        "01037CFB-8A89-46D7-B8FB-BD6D9B37B351": {
             "author": "com.apple.CoreML",
             "description": "CoreML Model Specification",
             "name": "model.mlmodel",
             "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "1D880E77-45D7-42F6-BE61-31CDD5476B86": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
         }
     },
-    "rootModelIdentifier": "505C3394-5B98-45F7-8F05-9C513B04AFCB"
+    "rootModelIdentifier": "01037CFB-8A89-46D7-B8FB-BD6D9B37B351"
 }