Adding turbo models (#967)

chidiwilliams · Nov 1, 2024 · 807b43f · 807b43f
1 parent 7650992
commit 807b43f
Show file tree

Hide file tree

Showing 4 changed files with 9 additions and 24 deletions.
diff --git a/buzz/model_loader.py b/buzz/model_loader.py
@@ -249,6 +249,7 @@ def get_local_model_path(self) -> Optional[str]:
     "large-v1": "7d99f41a10525d0206bddadd86760181fa920438b6b33237e3118ff6c83bb53d",
     "large-v2": "9a423fe4d40c82774b6af34115b8b935f34152246eb19e80e376071d3f999487",
     "large-v3": "64d182b440b98d5203c4f9bd541544d84c605196c4f7b845dfa11fb23594d1e2",
+    "large-v3-turbo": "1fc70f774d38eb169993ac391eea357ef47c88757ef72ee5943879b7e8e2bc69",
     "custom": None,
 }
 
@@ -383,19 +384,18 @@ def download_faster_whisper_model(
     size = model.whisper_model_size.to_faster_whisper_model_size()
     custom_repo_id = model.hugging_face_model_id
 
-    if size != WhisperModelSize.CUSTOM and size not in faster_whisper.utils._MODELS:
-        raise ValueError(
-            "Invalid model size '%s', expected one of: %s"
-            % (size, ", ".join(faster_whisper.utils._MODELS))
-        )
-
     if size == WhisperModelSize.CUSTOM and custom_repo_id == "":
         raise ValueError("Custom model id is not provided")
 
     if size == WhisperModelSize.CUSTOM:
         repo_id = custom_repo_id
     elif size == WhisperModelSize.LARGEV3:
         repo_id = "Systran/faster-whisper-large-v3"
+    # Maybe switch to 'mobiuslabsgmbh/faster-whisper-large-v3-turbo', seems to be used in
+    #  faster-whisper code https://github.com/SYSTRAN/faster-whisper/blob/master/faster_whisper/utils.py#L29
+    #  If so changes needed also in whisper_file_transcriber.py
+    elif size == WhisperModelSize.LARGEV3TURBO:
+        repo_id = "deepdml/faster-whisper-large-v3-turbo-ct2"
     else:
         repo_id = "guillaumekln/faster-whisper-%s" % size
 

diff --git a/buzz/transcriber/whisper_file_transcriber.py b/buzz/transcriber/whisper_file_transcriber.py
@@ -139,6 +139,8 @@ def transcribe_hugging_face(cls, task: FileTranscriptionTask) -> List[Segment]:
     def transcribe_faster_whisper(cls, task: FileTranscriptionTask) -> List[Segment]:
         if task.transcription_options.model.whisper_model_size == WhisperModelSize.CUSTOM:
             model_size_or_path = task.transcription_options.model.hugging_face_model_id
+        elif task.transcription_options.model.whisper_model_size == WhisperModelSize.LARGEV3TURBO:
+            model_size_or_path = "deepdml/faster-whisper-large-v3-turbo-ct2"
         else:
             model_size_or_path = task.transcription_options.model.whisper_model_size.to_faster_whisper_model_size()
 

diff --git a/buzz/widgets/preferences_dialog/models_preferences_widget.py b/buzz/widgets/preferences_dialog/models_preferences_widget.py
@@ -181,11 +181,6 @@ def reset(self):
                     model_size == WhisperModelSize.CUSTOM):
                 continue
 
-            # Skip turbo model for Faste Whisper and Whisper.cpp
-            if (self.model.model_type in {ModelType.FASTER_WHISPER, ModelType.WHISPER_CPP} and
-                    model_size == WhisperModelSize.LARGEV3TURBO):
-                continue
-
             model = TranscriptionModel(
                 model_type=self.model.model_type,
                 whisper_model_size=WhisperModelSize(model_size),

diff --git a/buzz/widgets/transcriber/transcription_options_group_box.py b/buzz/widgets/transcriber/transcription_options_group_box.py
@@ -52,7 +52,7 @@ def __init__(
 
         self.whisper_model_size_combo_box = QComboBox(self)
         self.whisper_model_size_combo_box.addItems(
-            [size.value.title() for size in WhisperModelSize if size not in {WhisperModelSize.CUSTOM, WhisperModelSize.LARGEV3TURBO}]
+            [size.value.title() for size in WhisperModelSize if size not in {WhisperModelSize.CUSTOM}]
         )
         self.whisper_model_size_combo_box.currentTextChanged.connect(
             self.on_whisper_model_size_changed
@@ -142,24 +142,12 @@ def reset_visible_rows(self):
                 and whisper_model_size == WhisperModelSize.CUSTOM),
             )
 
-        # Add turbo model size for whisper
-        if model_type == ModelType.WHISPER:
-            if self.whisper_model_size_combo_box.findText(WhisperModelSize.LARGEV3TURBO.value.title()) == -1:
-                self.whisper_model_size_combo_box.addItem(WhisperModelSize.LARGEV3TURBO.value.title())
-
         # Remove custom model size for whisper
         custom_model_index = (self.whisper_model_size_combo_box
                               .findText(WhisperModelSize.CUSTOM.value.title()))
         if model_type == ModelType.WHISPER and custom_model_index != -1:
             self.whisper_model_size_combo_box.removeItem(custom_model_index)
 
-        # Remove turbo model size for whisper_cpp and faster_whisper
-        turbo_model_index = (self.whisper_model_size_combo_box
-                             .findText(WhisperModelSize.LARGEV3TURBO.value.title()))
-        if ((model_type == ModelType.WHISPER_CPP or model_type == ModelType.FASTER_WHISPER)
-            and  turbo_model_index != -1):
-            self.whisper_model_size_combo_box.removeItem(turbo_model_index)
-
         # Add custom model size for whisper_cpp
         custom_model_index = (self.whisper_model_size_combo_box
                               .findText(WhisperModelSize.CUSTOM.value.title()))