Export whisper distil-large-v3 and distil-large-v3.5 to sherpa-onnx (#2506)

This PR adds support for exporting two new whisper distil models (distil-large-v3 and distil-large-v3.5) to ONNX format for use with sherpa-onnx. The changes enable these models to be processed through the existing export pipeline.

- Added support for distil-large-v3 and distil-large-v3.5 models in the export script
- Updated GitHub workflow to include the new models in the CI matrix
- Configured proper n_mels parameter (128) for the new distil models

@@ -16,8 +16,9 @@ jobs:
       fail-fast: false
       matrix:
         os: [macos-latest]
-        model: ["turbo", "distil-medium.en", "distil-small.en",  "tiny.en", "base.en", "small.en", "medium.en", "tiny", "base", "small", "medium", "medium-aishell", "large", "large-v1", "large-v2", "large-v3", "distil-large-v2"]
+        model: ["turbo", "distil-medium.en", "distil-small.en",  "tiny.en", "base.en", "small.en", "medium.en", "tiny", "base", "small", "medium", "medium-aishell", "large", "large-v1", "large-v2", "large-v3", "distil-large-v2", "distil-large-v3", "distil-large-v3.5"]
         # model: ["large", "large-v1", "large-v2", "large-v3", "distil-large-v2"]
+        # model: ["distil-large-v3.5", "distil-large-v3"]
         python-version: ["3.8"]
 
     steps:
@@ -47,6 +48,12 @@ jobs:
           elif [[ $model == distil-large-v2 ]]; then
             wget -q -O distil-large-v2-original-model.bin https://huggingface.co/distil-whisper/distil-large-v2/resolve/main/original-model.bin
             ls -lh
+          elif [[ $model == distil-large-v3 ]]; then
+            wget -q -O distil-large-v3-original-model.bin https://huggingface.co/distil-whisper/distil-large-v3-openai/resolve/main/model.bin
+            ls -lh
+          elif [[ $model == distil-large-v3.5 ]]; then
+            wget -q -O distil-large-v3.5-original-model.bin https://huggingface.co/distil-whisper/distil-large-v3.5-openai/resolve/main/model.bin
+            ls -lh
           elif [[ $model == distil-small.en ]]; then
             wget -q -O distil-small-en-original-model.bin https://huggingface.co/distil-whisper/distil-small.en/resolve/main/original-model.bin
             ls -lh
@@ -155,6 +162,7 @@ jobs:
 
           git status
           ls -lh
+          git lfs track "*.wav*"
           git lfs track "*onnx*"
           git lfs track "*weights*"
 

@@ -49,7 +49,8 @@ def get_args():
             "large-v1", "large-v2",
             "large", "large-v3", "turbo", # these three have feature dim 128
             "distil-medium.en", "distil-small.en", "distil-large-v2",
-            # "distil-large-v3", # distil-large-v3 is not supported!
+            "distil-large-v3",
+            "distil-large-v3.5",
             # for fine-tuned models from icefall
             "medium-aishell",
             ],
@@ -348,6 +349,32 @@ def main():
             """
             )
         model = whisper.load_model(filename)
+    elif name == "distil-large-v3":
+        filename = "./distil-large-v3-original-model.bin"
+        if not Path(filename).is_file():
+            raise ValueError(
+                """
+                Please go to https://huggingface.co/distil-whisper/distil-large-v3-openai
+                to download model.bin
+                You can use the following command to do that:
+
+                wget -O distil-large-v3-original-model.bin https://huggingface.co/distil-whisper/distil-large-v3-openai/resolve/main/model.bin
+            """
+            )
+        model = whisper.load_model(filename)
+    elif name == "distil-large-v3.5":
+        filename = "./distil-large-v3.5-original-model.bin"
+        if not Path(filename).is_file():
+            raise ValueError(
+                """
+                Please go to https://huggingface.co/distil-whisper/distil-large-v3.5-openai/
+                to download model.bin
+                You can use the following command to do that:
+
+                wget -O distil-large-v3.5-original-model.bin https://huggingface.co/distil-whisper/distil-large-v3.5-openai/resolve/main/model.bin
+            """
+            )
+        model = whisper.load_model(filename)
     elif name == "distil-small.en":
         filename = "./distil-small-en-original-model.bin"
         if not Path(filename).is_file():
@@ -405,10 +432,17 @@ def main():
     audio = whisper.pad_or_trim(audio)
     assert audio.shape == (16000 * 30,), audio.shape
 
-    if args.model in ("large", "large-v3", "turbo"):
+    if args.model in ("distil-large-v3", "distil-large-v3.5"):
+        n_mels = 128
+    elif args.model in (
+        "large",
+        "large-v3",
+        "turbo",
+    ):
         n_mels = 128
     else:
         n_mels = 80
+
     mel = (
         whisper.log_mel_spectrogram(audio, n_mels=n_mels).to(model.device).unsqueeze(0)
     )