Upload OpenVoice V2 Core ML packages (ToneColorConverter split + E2E)

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93546068a0df4e38290f34c24c24084d67981e63d83cde98e25b70e9479189ab
+size 127132

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cdc47ffb85ccdc8ddd3bbee4f4301112a53565e5720ba5c3c732f21f7c2e09e
+size 28928128

ToneColorConverter_Decoder.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "31187191-7E33-4AB3-94DB-5394DBB2EF43": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "EEC631F0-A5E6-4F3A-9F19-3EB106D52DB3": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        }
+    },
+    "rootModelIdentifier": "31187191-7E33-4AB3-94DB-5394DBB2EF43"
+}

ToneColorConverter_E2E.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1da0f0ed72c1007d9bb2d6c58ae37107bd7f4e26a65da7dabfd2ea20232f988d
+size 484140

ToneColorConverter_E2E.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08cf5ec87cb39715a545d90b4ba8dc1d15109e180d499c913c4dd5d2939e5a1a
+size 63887808

ToneColorConverter_E2E.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "60A425C0-2DFD-487F-9628-3E99AB50B638": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "6C23D1FD-1ACF-4814-982A-4B6733963FA3": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        }
+    },
+    "rootModelIdentifier": "60A425C0-2DFD-487F-9628-3E99AB50B638"
+}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:663b25b79f95e54e63ed859b448b21e1169500c716cf6aa042d483d03218d9ff
+size 127591

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae50e6ff9bf98338386b8f5d96d0d8028c92017ac8fe6d1dfe715286a456cc60
+size 17344832

ToneColorConverter_Flow_Forward.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "21212833-7C2B-4360-9B32-98AEE50CA5DD": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "423DEA45-73A3-40B2-A1A3-95C38F5F026C": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "423DEA45-73A3-40B2-A1A3-95C38F5F026C"
+}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39361657aec3a0c5bf09e4c5b78275d1d98f0d72e28573b20f7f0f7748d9e4a6
+size 128821

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bca8988c3ab9e39cc328f22595f8811733a0b9ad029a874d28108ffd6d64cedf
+size 17344832

ToneColorConverter_Flow_Reverse.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "B7142C0B-E6F1-4DA2-811D-3F592A214E49": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "BFEB036E-9797-4060-A53D-6231E46967D3": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "BFEB036E-9797-4060-A53D-6231E46967D3"
+}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa616848c8b3802647d7003303626e0c56ed2ed800b2524a9260c5ed524aea7a
+size 115188

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35ab1cef640f0ec3b9ea4df7a15754d8c81a47d9db1c6c1129fa440ce401982e
+size 17614400

ToneColorConverter_PosteriorEncoder.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "1F1AA866-CE50-418F-BF1B-1B637C476D10": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "BC7DD43C-A293-405F-A0A6-7F61C9BAF012": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "BC7DD43C-A293-405F-A0A6-7F61C9BAF012"
+}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87711f3ea0c492ede0545ed9d59a56d0df59718b35bfda234a708315c6edae14
+size 26635

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b50c4ca00b72862f7cd974f9bef7d1dd36d2a86adc0b6b626116fb26b5cc6de
+size 1627840

ToneColorConverter_ReferenceEncoder.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "286603A1-1E6C-41C7-BB30-93AE1B8D3A54": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        },
+        "40E2CAEC-F117-4137-BF54-9B3A9DADBC0C": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        }
+    },
+    "rootModelIdentifier": "40E2CAEC-F117-4137-BF54-9B3A9DADBC0C"
+}

pipeline_info.json

@@ -0,0 +1,92 @@
+{
+  "description": "OpenVoice V2 ToneColorConverter を5つの CoreML モデルに分割。voice_conversion パスを再現する実行順序は以下の通り。",
+  "pipeline": [
+    {
+      "step": 1,
+      "model": "ToneColorConverter_ReferenceEncoder.mlpackage",
+      "role": "参照音声の mel-spectrogram → tone color embedding v(C)",
+      "inputs": {
+        "spec": "[1, T, n_mels] float32"
+      },
+      "outputs": {
+        "g": "[1, gin_channels, 1] float32"
+      }
+    },
+    {
+      "step": 2,
+      "model": "ToneColorConverter_PosteriorEncoder.mlpackage",
+      "role": "ベース音声の STFT スペクトル → 潜在変数 z",
+      "inputs": {
+        "spec": "[1, n_fft, T] float32",
+        "spec_lengths": "[1] int32",
+        "g": "[1, gin_channels, 1] float32  (ソーススピーカーの g)"
+      },
+      "outputs": {
+        "z": "[1, inter_channels, T] float32",
+        "y_mask": "[1, 1, T] float32"
+      }
+    },
+    {
+      "step": 3,
+      "model": "ToneColorConverter_Flow_Forward.mlpackage",
+      "role": "z → z_p (ソーストーンカラー除去)",
+      "inputs": {
+        "z": "step2.z",
+        "y_mask": "step2.y_mask",
+        "g": "ソース g"
+      },
+      "outputs": {
+        "z_p": "[1, inter_channels, T] float32"
+      }
+    },
+    {
+      "step": 4,
+      "model": "ToneColorConverter_Flow_Reverse.mlpackage",
+      "role": "z_p → z_hat (ターゲットトーンカラー注入)",
+      "inputs": {
+        "z_p": "step3.z_p",
+        "y_mask": "step2.y_mask",
+        "g_tgt": "ターゲット g"
+      },
+      "outputs": {
+        "z_hat": "[1, inter_channels, T] float32"
+      }
+    },
+    {
+      "step": 5,
+      "model": "ToneColorConverter_Decoder.mlpackage",
+      "role": "z_hat → 波形 (HiFi-GAN)",
+      "inputs": {
+        "z_hat": "step4.z_hat",
+        "y_mask": "step2.y_mask",
+        "g": "ターゲット g"
+      },
+      "outputs": {
+        "audio": "[1, 1, T_audio] float32"
+      }
+    }
+  ],
+  "hyperparameters": {
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "n_fft": 513,
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "gin_channels": 256,
+    "upsample_rates": [
+      8,
+      8,
+      2,
+      2
+    ]
+  },
+  "notes": [
+    "MeloTTS (BaseSpeakerTTS) は可変長 tokenizer のため CoreML 化対象外。",
+    "ベース音声は Python/MeloTTS で生成し、そのスペクトルを step2 に渡す。",
+    "PosteriorEncoder は Core ML 用に後験平均 (tau=0 相当) のみ。PyTorch の randn サンプリングは含まない。",
+    "tone color embedding は事前計算して .pth/.bin で保存しておくと効率的。",
+    "iOS 推論時は AVAudioEngine で sampling_rate に合わせてリサンプリング要。"
+  ]
+}

pipeline_info_e2e.json

@@ -0,0 +1,16 @@
+{
+  "model": "ToneColorConverter_E2E.mlpackage",
+  "inputs": {
+    "y": "[1, 513, T] float32 — linear STFT magnitude (same layout as PyTorch OpenVoice)",
+    "y_lengths": "[1] int32",
+    "sid_src": "[1, 256, 1] float32 — source tone embedding",
+    "sid_tgt": "[1, 256, 1] float32 — target tone embedding"
+  },
+  "outputs": {
+    "audio_out": "[1, 1, T_audio] float32"
+  },
+  "notes": [
+    "tau=0 固定（後験からのサンプリングなし）。",
+    "5 分割版と異なり中間 z を取り出せないが呼び出しは1回で済む。"
+  ]
+}