TTS Client

File: main_logic/tts_client.py (~2300 lines)

The TTS client handles text-to-speech synthesis across multiple providers with a unified queue-based interface.

Factory function

from main_logic.tts_client import get_tts_worker

worker = get_tts_worker(config)

Creates a TTS worker configured for the active provider and voice settings.

Supported providers

Provider	Module	Features
DashScope CosyVoice	Cloud	High quality, voice cloning, streaming
DashScope TTS V2	Cloud	Lower latency variant
GPT-SoVITS	Local	Fully offline, customizable
Custom	HTTP	Any OpenAI-compatible TTS endpoint

Queue architecture

The TTS client uses a producer-consumer pattern:

1. Request queue: Text sentences enqueued by the session manager
2. Worker thread: Dequeues text, calls the TTS API, produces audio chunks
3. Response queue: Audio chunks ready for resampling and WebSocket delivery

Voice cloning flow

1. User uploads audio sample via /api/characters/voice_clone
2. Audio is sent to DashScope's voice enrollment API
3. A voice_id is returned and stored in character config
4. Subsequent TTS calls include the voice_id for personalized synthesis

Interruption

When the user interrupts:

1. Both queues are flushed
2. Any in-progress TTS API call is cancelled
3. The worker is immediately ready for new input