Skip to content

Qwen Streaming TTS

Synthesizes incremental response text into streaming speech.

Property Value
Node type qwen.tts_realtime
Layer / kind algorithm / transform
Capability speech.tts.cancellable_streaming

Ports

Port Direction Schema
text_in Input Text Incremental synthesis text
signal_in Input Signal Closes the WebSocket, clears queues, and rejects old-turn Text
audio_out Output Audio PCM S16LE, 24 kHz, mono, streaming

Configuration

model defaults to qwen3-tts-flash-realtime; voice defaults to Cherry; language_type defaults to Auto. The worker reuses one TTS session across sentence chunks to avoid reconnect gaps. A Signal closes that session, advances its generation, and rejects late PCM. It also records the cancellation sequence and rejects stale Text. The worker requests Runtime-managed internal wakeups through its Context, so no Clock Node is required. max_results_per_wakeup defaults to 64. Demo 2 resamples 24 kHz output to Agora's 48 kHz PCM.

See Alibaba Cloud's Qwen real-time speech synthesis protocol.