Docker container reload when hitting generate speech for recorded voice #337
Description
Environment Details
- OS: macOS (MacBook Pro M3)
- Memory: 36GB RAM
- Docker Desktop: Running in containerized mode (Apple Silicon)
- Working Model: VoiceBox, Kokoro-82M (using existing voices in the app)
- Ram 16GB configured in the docker-compose file to be used.
Description of Issue
The application runs successfully when generating speech using pre-existing voice embeddings and the Kokoro-82M model. However, attempting to use recorded audio file with models like Qwen or TADA causes the Docker container to crash/reload, resulting in a failed generation process.
Steps to Reproduce
- Start the Docker container on MacBook Pro M3 (36GB RAM).
- Load a pre-trained voice model (e.g., Kokoro-82M) and generate audio successfully.
- Attempt to load a custom recorded audio file for cloning.
- Switch or initialize the Qwen or TADA model.
- Trigger generation.
Expected Behavior
The system should accept the uploaded audio, load the specified model (Qwen/TADA), and generate the cloned voice without interrupting the container process.
Actual Behavior
The Docker container restarts immediately after the model load successfully and start the process.
Generation fails with no successful output.
No specific error message is returned in the UI/logs before the crash (container just reload immeditaly).
