- Add SileroTTS service with GPU/CPU support
- Update TTS API to support configurable engines (silero/edge)
- Add TTS_ENGINE and SILERO_SPEAKER config options
- Set Silero as default TTS engine
- Update requirements.txt with silero dependency
- Replace broken import from inference import load_model with SeedVCWrapper
- SeedVCWrapper properly loads all models (DiT, Whisper, BigVGAN, CAMPPlus, RMVPE)
- Add support for F0 conditioning and voice conversion parameters
- Implement proper voice conversion using wrapper.convert_voice()
- Keep fallback to passthrough mode if wrapper fails to load
Fixes: Could not import Seed-VC modules: cannot import name 'load_model' from 'inference'
- Add proper error handling around pydub conversion
- Use absolute paths instead of relative paths
- Ensure output directory exists before saving
- Verify file was actually created before returning
- Add fallback to MP3 if conversion fails
- Update tts.py to handle both .wav and .mp3 extensions
- Add detailed logging for debugging
- Use pydub to convert MP3 to WAV in passthrough mode
- Fixed fallback conversion when Seed-VC is not available
- Output WAV format as expected by the API
