Spaces:
Paused
Paused
Commit History
-
Fix TTS language issue by dynamically updating model in /translate-audio endpoint
Verified signature2b35bda
-
Removed webrtcvad to avoid the build error.
Verified signature49a0a85 -
Replaced librosa with torchaudio for audio loading and resampling. Added speech detection (energy-based or webrtcvad for accuracy). Improved /translate-audio endpoint to handle silent audio gracefully.
Verified signatureb0c2331 -
Updated title to Talklas API. Added description of /translate-audio endpoint with speech detection feature. Included Features section (speech detection, supported models) and Supported Languages section.
Verified signature8157595 -
Added apt-get install for libsndfile1 (required by soundfile) and ffmpeg (required by torchaudio).
Verified signature04d442c -
Replaced librosa with torchaudio; removed librosa and numba. Added torchaudio (audio processing) and webrtcvad (optional voice activity detection).
Verified signature40a94e0 -
Pin librosa to a known stable version and ensure numba is compatible
Verified signature824f995
-
Fix TTS error by updating save_pcm_to_wav to handle 16-bit audio samples correctly using NumPy tobytes()
Verified signaturebf59705 -
Fix PermissionError by changing AUDIO_DIR to /tmp/audio_output for Hugging Face Spaces
Verified signaturee79c9c5 -
Fix PermissionError by setting AUDIO_DIR to /app/audio_output for Hugging Face Spaces compatibility
Verified signature060630c -
Update audio handling: Save synthesized audio as WAV, return URL, set 10-min expiration with 5-min cleanup interval
Verified signatureeacfbe2
-
Switch STT model to facebook/mms-1b-all with Whisper fallback
Verified signature98d2781 -
Fix STT error by removing return_attention_mask and address forced_decoder_ids warning
Verified signature62f9c95 -
Fix MT error by removing clean_up_tokenization_spaces and address STT warnings
Verified signaturee656c37 -
Update app.py
Verified signature989a3f5 -
Add MT with NLLB-200-distilled-600M and TTS with MMS-TTS
Verified signaturef56283f -
Fix torch import error in translate-audio endpoint
Verified signaturef8dca01 -
Add STT functionality with openai/whisper-tiny
Verified signaturefaf4aa8 -
Add STT functionality with openai/whisper-tiny
Verified signatureec27d4d -
Update README.md
Verified signaturec10c930 -
Update app.py
Verified signatureaef3d4f -
Deploy minimal FastAPI app to test Hugging Face Spaces compatibility
Verified signaturef5b1ec8 -
Deploy minimal FastAPI app to test Hugging Face Spaces compatibility
Verified signature5f26e89 -
Update app.py
Verified signature2fe79f5 -
Added pydantic for speech translation system
Verified signature55878bb -
Update Dockerfile with proper caching and environment variables
Verified signature4a3f84d -
Implement lazy loading of ML models to fix startup timeout on HF Spaces
Verified signature352aa0e -
Fix MT model loading: Revert to nllb-200-distilled-600M and add fallback
Verified signature- Reverted translation model to facebook/nllb-200-distilled-600M (facebook/nllb-200-distilled-200M does not exist) - Added fallback mechanism in _initialize_mt_model to handle model loading failures - Modified translate_text to return source text if the translation model is not loaded - Added initialization of mt_model and mt_tokenizer as None in __init__6a1bf6c -
Further optimize memory usage: Switch to nllb-200-distilled-200M and add Uvicorn debug logging
Verified signature- Switched translation model from facebook/nllb-200-distilled-600M to facebook/nllb-200-distilled-200M to reduce memory usage - Added debug logging around Uvicorn startup to diagnose startup issues - Added debug print statements in _initialize_tts_model for better logginge868e42 -
Optimize app for Hugging Face Spaces: Switch to whisper-tiny, add health check, and improve model loading
Verified signature- Switched STT model from openai/whisper-small to openai/whisper-tiny to reduce memory usage and prevent startup timeout - Optimized update_languages method to only reload TTS model, not STT model - Added debug logging in __init__ to confirm model loading - Added /health endpoint for Hugging Face Spaces health check - Fixed clean_up_tokenization_spaces warning in _initialize_mt_model by setting clean_up_tokenization_spaces=True1881cc0 -
Replaced the openai/whisper-tiny references with openai/whisper-small
Verified signature4279e53 -
Modified the _initialize_tts_model method to include the clean_up_tokenization_spaces parameter; Added logging configuration to configure the logging level for transformers in app.py
Verified signaturef12e9dc -
Modified the tokenization step to include clean_up_tokenization_spaces=True; Added clean_up_tokenization_spaces=True in the text_to_speech method; Added a print statement to confirm the TTS model is loaded
Verified signature5b5fc47 -
Removed the TRANSFORMERS_CACHE environment variable
Verified signature3cc49a2 -
Removed the TRANSFORMERS_CACHE environment variable and its associated debug print statement.
Verified signature741c20d -
Added python-multipart
Verified signaturefa5448d -
Update app.py
Verified signature52830c5 -
Set the Cache Directory Explicitly
Verified signaturede6eef4 -
Update requirements.txt
Verified signature43a7f05 -
Create Dockerfile
Verified signaturecb67677 -
Set sdk: docker
Verified signature7e11802 -
Removed SDK
Verified signature77a6466 -
Update requirements.txt
Verified signature388f530 -
Removed flask, and flask-cors; Added fastapi, and uvicorn
Verified signaturefba973c -
Modified the code with API endpoints
Verified signature4763326 -
Update app.py
Verified signaturefff0177 -
Update README.md
Verified signature6b19945 -
Added flask, and flask-cors; Removed fastapi, uvicorn, and pydantic as dependencies
Verified signature860aa9c -
API integration for the Talklas pipeline
Verified signaturede2c35e
-
Added fastapi, uvicorn, and pydantic as dependencies
Verified signaturef24e508