Added noise suppression and fixed whisper v3 issues

tmoroney · Nov 26, 2024 · e9ebfe9 · e9ebfe9
1 parent 9fa13b0
commit e9ebfe9
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 7 deletions.
diff --git a/AutoSubs-App/src/GlobalContext.tsx b/AutoSubs-App/src/GlobalContext.tsx
@@ -94,7 +94,7 @@ export function GlobalProvider({ children }: React.PropsWithChildren<{}>) {
     const [translate, setTranslate] = useState(false);
     const [diarize, setDiarize] = useState(false);
     const [maxWords, setMaxWords] = useState(6);
-    const [maxChars, setMaxChars] = useState(30);
+    const [maxChars, setMaxChars] = useState(25);
     const [markIn, setMarkIn] = useState(0);
 
     async function getFullTranscriptPath() {
@@ -149,7 +149,7 @@ export function GlobalProvider({ children }: React.PropsWithChildren<{}>) {
             setTranslate(result.storedTranslate || false);
             setDiarize(result.storedDiarize || false);
             setMaxWords(result.storedMaxWords || 6);
-            setMaxChars(result.storedMaxChars || 30);
+            setMaxChars(result.storedMaxChars || 25);
         }).catch((error) => {
             console.error("Error initializing state:", error);
         });

diff --git a/Mac-Server/transcription-server.py b/Mac-Server/transcription-server.py
@@ -76,15 +76,15 @@ def __getattr__(self, attr):
     "base": "mlx-community/whisper-base-mlx-q4",
     "small": "mlx-community/whisper-small-mlx",
     "medium": "mlx-community/whisper-medium-mlx",
-    "large": "mlx-community/whisper-large-v3-turbo",
+    "large": "mlx-community/distil-whisper-large-v3",
 }
 
 english_models = {
     "tiny": "mlx-community/whisper-tiny.en-mlx",
     "base": "mlx-community/whisper-base.en-mlx",
     "small": "mlx-community/whisper-small.en-mlx",
     "medium": "mlx-community/whisper-medium.en-mlx",
-    "large": "mlx-community/whisper-large-v3-turbo",
+    "large": "mlx-community/distil-whisper-large-v3",
 }
 
 def is_model_cached_locally(model_id, revision=None):
@@ -152,14 +152,13 @@ def inference(audio, **kwargs) -> dict:
             language=kwargs["language"],
             verbose=True,
             task=kwargs["task"]
+
         )
-    # Ensure segments are sorted
-    output["segments"] = sorted(output["segments"], key=lambda x: x["start"])
     return output
 
 def transcribe_audio(audio_file, kwargs, max_words, max_chars):
     print("Starting transcription...")
-    whisperResult = stable_whisper.transcribe_any(inference, audio_file, inference_kwargs = kwargs, vad=False)
+    whisperResult = stable_whisper.transcribe_any(inference, audio_file, inference_kwargs = kwargs, vad=True, only_voice_freq=True, force_order=True)
     whisperResult.split_by_length(max_words=max_words, max_chars=max_chars)
     return whisperResult.to_dict()