How eepstein201/qwen3-tts-advanced-eme works
Qwen3-TTS is a local voice-generation product: users submit text and a voice mode through the CLI, Python/client API, or Gradio UI; the FastAPI server validates requests, loads the requested model, synthesizes audio, and returns completed or streamed results. Clone, Design, and Custom share server-side model management and the inference engine. The README also documents prompt creation, transcription, chunked playback, history/model controls, and Torch, MLX, and optional vLLM backends. Paths for several capabilities are supported by the tree and README, but their internal wiring was not sampled; the graph omits uncertain links.
Main entry points: app.py FastAPI server · inference.py Inference dispatch · generator.py Server client · app_generation.py Generation routes
Components 24
User interfaces
- CLI
qwen3_tts/cli.py - Batch and dialogue
qwen3_tts/interface/cli - Python generation
qwen3_tts/interface/generate.py - Gradio UI
qwen3_tts/interface/ui - Server client
qwen3_tts/server/client/generator.py
Server orchestration
- Streaming endpoint
qwen3_tts/server/websocket.py - FastAPI server
qwen3_tts/server/app.py - Generation routes
qwen3_tts/server/app_generation.py - Model routes
qwen3_tts/server/app_models.py - Model loading
qwen3_tts/server/model_loading.py - Prompt routes
qwen3_tts/server/app_prompts.py - Prompt loading
qwen3_tts/server/prompt_loading.py - Request validation
qwen3_tts/server/validation.py
Synthesis engine
- Inference dispatch
qwen3_tts/core/engine/inference.py - Text processing
qwen3_tts/core/engine/text_processing.py - Audio processing
qwen3_tts/core/engine/audio_processing.py - Speech transcription
qwen3_tts/core/engine/asr.py
Configuration and voices
- Runtime configuration
qwen3_tts/core/config - Voice prompts
qwen3_tts/core/engine/voice_prompt.py
Inference backends
- Torch and MLX
qwen3_tts/core/engine/model_loader.py - vLLM client
qwen3_tts/server/vllm_client.py - vLLM adapter
qwen3_tts/core/engine_vllm.py
Other components
- User
- Generated audio
Connections 27
- User → CLI: selects workflow
- User → Gradio UI: submits request
- User → Python generation: calls API
- CLI → Server client: requests generation
- Batch and dialogue → Server client: requests generation
- Python generation → Server client: requests generation
- Gradio UI → Server client: calls server
- Server client → FastAPI server: HTTP requests
- FastAPI server → Request validation: validates input
- FastAPI server → Generation routes: dispatches routes
- Generation routes → Inference dispatch: runs synthesis
- Inference dispatch → Text processing: prepares text
- Inference dispatch → Audio processing: processes audio
- Inference dispatch → Torch and MLX: uses backend
- Inference dispatch → Voice prompts: uses voice prompt
- Inference dispatch → Runtime configuration: reads settings
- Model routes → Model loading: loads models
- FastAPI server → Model routes: dispatches routes
- FastAPI server → Prompt routes: dispatches routes
- Prompt routes → Prompt loading: loads prompts
- Prompt loading → Voice prompts: uses prompt handling
- Generation routes → Generated audio: returns audio
- Generation routes → Streaming endpoint: streams chunks
- Streaming endpoint → Generated audio: streams audio
- Speech transcription → Generated audio: transcribes audio
- FastAPI server → vLLM client: uses when configured
- vLLM client → vLLM adapter: calls local backend