程序包 io.agentscope.core.model.tts

接口 RealtimeTTSModel

所有超级接口:
TTSModel
所有已知实现类:
DashScopeRealtimeTTSModel
public interface RealtimeTTSModel extends TTSModel
Interface for real-time TTS models that support streaming input.

This interface extends TTSModel with methods for streaming text input, enabling "speak as you generate" functionality. Unlike the base TTSModel which requires complete text upfront, this interface allows pushing text incrementally while maintaining context continuity.

Key difference from TTSModel:

  • TTSModel: One-time input, streaming output (e.g., HTTP + SSE)
  • RealtimeTTSModel: Streaming input + streaming output (e.g., WebSocket)

Typical models:

  • qwen3-tts-flash-realtime - WebSocket real-time model
  • cosyvoice-v2 - WebSocket streaming model

Usage Example: 


 RealtimeTTSModel tts = DashScopeRealtimeTTSModel.builder()
     .apiKey(apiKey)
     .modelName("qwen3-tts-flash-realtime")
     .voice("Cherry")
     .build();

 // Start streaming session
 tts.startSession();

 // Push text chunks (context maintained for natural prosody)
 tts.push("Hello, ").subscribe(audio -> player.play(audio));
 tts.push("welcome to ").subscribe(audio -> player.play(audio));
 tts.push("AgentScope.").subscribe(audio -> player.play(audio));

 // Finish session and get remaining audio
 tts.finish().blockLast();

 // Clean up
 tts.close();
另请参阅:

  • 方法概要

    修饰符和类型
    方法
    说明
    void
    close()
    Closes the TTS session and releases resources.
    reactor.core.publisher.Flux<AudioBlock>
    finish()
    Signals end of input and flushes remaining audio.
    reactor.core.publisher.Flux<AudioBlock>
    getAudioStream()
    Gets the audio stream for receiving synthesized audio.
    reactor.core.publisher.Flux<AudioBlock>
    push(String text)
    Pushes text incrementally to the TTS service.
    void
    startSession()
    Starts a new streaming session.
    reactor.core.publisher.Flux<AudioBlock>
    synthesizeStream(String text)
    Synthesizes text using streaming and returns audio blocks.

    从接口继承的方法 io.agentscope.core.model.tts.TTSModel

    getModelName, synthesize

  • 方法详细资料

    • startSession

      void startSession()
      Starts a new streaming session.

      This typically establishes a WebSocket connection to the TTS service. Must be called before push(String) or finish().

      抛出:
      TTSException - if session cannot be started

    • push

      reactor.core.publisher.Flux<AudioBlock> push(String text)
      Pushes text incrementally to the TTS service.

      Text is buffered and synthesized while maintaining context continuity, resulting in natural prosody and intonation across chunks.

      参数:
      text - the text chunk to synthesize
      返回:
      Flux of AudioBlock containing synthesized audio

    • finish

      reactor.core.publisher.Flux<AudioBlock> finish()
      Signals end of input and flushes remaining audio.

      Call this when all text has been pushed to receive any remaining synthesized audio.

      返回:
      Flux of AudioBlock containing remaining audio

    • synthesizeStream

      reactor.core.publisher.Flux<AudioBlock> synthesizeStream(String text)
      Synthesizes text using streaming and returns audio blocks.

      This is a convenience method that handles the full session lifecycle (start, push, finish) for a single text input.

      参数:
      text - the complete text to synthesize
      返回:
      Flux of AudioBlock as audio is synthesized

    • close

      void close()
      Closes the TTS session and releases resources.

      Should be called when the model is no longer needed to close WebSocket connections and clean up resources.

    • getAudioStream

      reactor.core.publisher.Flux<AudioBlock> getAudioStream()
      Gets the audio stream for receiving synthesized audio.

      This method returns a Flux that emits audio blocks as they are synthesized. Subscribe to this stream once after calling startSession() to receive audio data.

      Important: Only subscribe once per session. Multiple subscriptions may cause duplicate audio playback.

      返回:
      Flux of AudioBlock that emits audio as it's synthesized