change transcribe mechanism when uploading audio

Dockerfile

@@ -22,4 +22,4 @@ RUN pip install --no-cache-dir --upgrade -r requirements.txt
 
 COPY --chown=user . /app
 # CMD ["gunicorn", "-b", "0.0.0.0:7860", "app:app"]
-CMD ["gunicorn", "--timeout", "120", "-b", "0.0.0.0:7860", "app:app"]
+CMD ["gunicorn", "--timeout", "300", "--workers", "1", "--threads", "2", "-b", "0.0.0.0:7860", "app:app"]

app.py

@@ -68,15 +68,18 @@ pipe = pipeline(
     model=whisper_model,
     tokenizer=processor.tokenizer,
     feature_extractor=processor.feature_extractor,
-    chunk_length_s=10,
-    batch_size=4,  # batch size for inference - set based on your device
+    chunk_length_s=30,
+    batch_size=1,  # batch size for inference - set based on your device
     torch_dtype=torch_dtype,
     device=device,
+    max_new_tokens=128,  # Limit text generation
+    return_timestamps=False,  # Save memory
+    device_map="auto",    # Better resource handling
 )
 
 
 def whisper_api(temp_audio_path):
-    result = pipe(temp_audio_path, return_timestamps=False, generate_kwargs={"language": "indonesian"})
+    result = pipe(temp_audio_path, generate_kwargs={"language": "indonesian", "task": "transcribe"})
     print(result["text"])
     return result