Alibaba Cloud Model Studio:Real-time speech synthesis

Qwen-TTS

1. Crie o cliente

   Python

   Crie um arquivo Python chamado tts_realtime_client.py e copie o código a seguir para o arquivo:

   # -- coding: utf-8 --
   
   import asyncio
   import websockets
   import json
   import base64
   import time
   from typing import Optional, Callable, Dict, Any
   from enum import Enum
   
   class SessionMode(Enum):
       SERVER_COMMIT = "server_commit"
       COMMIT = "commit"
   
   class TTSRealtimeClient:
       """
       Client for interacting with the TTS Realtime API.
   
       This class provides methods for connecting to the TTS Realtime API, sending text data,
       receiving audio output, and managing WebSocket connections.
   
       Attributes:
           base_url (str):
               Base URL of the Realtime API.
           api_key (str):
               API Key for authentication.
           voice (str):
               Voice used for server-side speech synthesis.
           mode (SessionMode):
               Session mode, either server_commit or commit.
           audio_callback (Callable[[bytes], None]):
               Callback function for receiving audio data.
           language_type(str)
               Language for synthesized speech. Options: Chinese, English, German, Italian, Portuguese, Spanish, Japanese, Korean, French, Russian, Auto
       """
   
       def __init__(
               self,
               base_url: str,
               api_key: str,
               voice: str = "Cherry",
               mode: SessionMode = SessionMode.SERVER_COMMIT,
               audio_callback: Optional[Callable[[bytes], None]] = None,
           language_type: str = "Auto"):
           self.base_url = base_url
           self.api_key = api_key
           self.voice = voice
           self.mode = mode
           self.ws = None
           self.audio_callback = audio_callback
           self.language_type = language_type
   
           # Current response state
           self._current_response_id = None
           self._current_item_id = None
           self._is_responding = False
           self._response_done_future = None
   
       async def connect(self) -> None:
           """Establish WebSocket connection with the TTS Realtime API."""
           headers = {
               "Authorization": f"Bearer {self.api_key}"
           }
   
           self.ws = await websockets.connect(self.base_url, additional_headers=headers)
   
           # Set default session configuration
           await self.update_session({
               "mode": self.mode.value,
               "voice": self.voice,
               # To use the instruction control feature, uncomment the lines below and replace the model with qwen3-tts-instruct-flash-realtime in server_commit.py or commit.py
               # "instructions": "Speak quickly with a noticeable rising intonation, suitable for introducing fashion products.",
               # "optimize_instructions": true
               "language_type": self.language_type,
               "response_format": "pcm",
               "sample_rate": 24000
           })
   
       async def send_event(self, event) -> None:
           """Send an event to the server."""
           event['event_id'] = "event_" + str(int(time.time() * 1000))
           print(f"Sending event: type={event['type']}, event_id={event['event_id']}")
           await self.ws.send(json.dumps(event))
   
       async def update_session(self, config: Dict[str, Any]) -> None:
           """Update session configuration."""
           event = {
               "type": "session.update",
               "session": config
           }
           print("Updating session configuration: ", event)
           await self.send_event(event)
   
       async def append_text(self, text: str) -> None:
           """Send text data to the API."""
           event = {
               "type": "input_text_buffer.append",
               "text": text
           }
           await self.send_event(event)
   
       async def commit_text_buffer(self) -> None:
           """Commit text buffer to trigger processing."""
           event = {
               "type": "input_text_buffer.commit"
           }
           await self.send_event(event)
   
       async def clear_text_buffer(self) -> None:
           """Clear the text buffer."""
           event = {
               "type": "input_text_buffer.clear"
           }
           await self.send_event(event)
   
       async def finish_session(self) -> None:
           """End the session."""
           event = {
               "type": "session.finish"
           }
           await self.send_event(event)
   
       async def wait_for_response_done(self):
           """Wait for the response.done event"""
           if self._response_done_future:
               await self._response_done_future
   
       async def handle_messages(self) -> None:
           """Handle messages from the server."""
           try:
               async for message in self.ws:
                   event = json.loads(message)
                   event_type = event.get("type")
   
                   if event_type != "response.audio.delta":
                       print(f"Received event: {event_type}")
   
                   if event_type == "error":
                       print("Error: ", event.get('error', {}))
                       continue
                   elif event_type == "session.created":
                       print("Session created, ID: ", event.get('session', {}).get('id'))
                   elif event_type == "session.updated":
                       print("Session updated, ID: ", event.get('session', {}).get('id'))
                   elif event_type == "input_text_buffer.committed":
                       print("Text buffer committed, item ID: ", event.get('item_id'))
                   elif event_type == "input_text_buffer.cleared":
                       print("Text buffer cleared")
                   elif event_type == "response.created":
                       self._current_response_id = event.get("response", {}).get("id")
                       self._is_responding = True
                       # Create a new future to wait for response.done
                       self._response_done_future = asyncio.Future()
                       print("Response created, ID: ", self._current_response_id)
                   elif event_type == "response.output_item.added":
                       self._current_item_id = event.get("item", {}).get("id")
                       print("Output item added, ID: ", self._current_item_id)
                   # Handle audio delta
                   elif event_type == "response.audio.delta" and self.audio_callback:
                       audio_bytes = base64.b64decode(event.get("delta", ""))
                       self.audio_callback(audio_bytes)
                   elif event_type == "response.audio.done":
                       print("Audio generation completed")
                   elif event_type == "response.done":
                       self._is_responding = False
                       self._current_response_id = None
                       self._current_item_id = None
                       # Mark future as done
                       if self._response_done_future and not self._response_done_future.done():
                           self._response_done_future.set_result(True)
                       print("Response completed")
                   elif event_type == "session.finished":
                       print("Session finished")
   
           except websockets.exceptions.ConnectionClosed:
               print("Connection closed")
           except Exception as e:
               print("Error handling messages: ", str(e))
   
       async def close(self) -> None:
           """Close the WebSocket connection."""
           if self.ws:
               await self.ws.close()
   

   Java

   Crie um arquivo Java chamado TTSRealtimeClient.java e copie o código abaixo para o arquivo:

   import com.google.gson.Gson;
   import com.google.gson.JsonObject;
   import org.java_websocket.client.WebSocketClient;
   import org.java_websocket.handshake.ServerHandshake;
   
   import java.net.URI;
   import java.util.Base64;
   import java.util.HashMap;
   import java.util.Map;
   import java.util.concurrent.CountDownLatch;
   import java.util.function.Consumer;
   
   /**
    * A client for interacting with the TTS Realtime API.
    *
    * This class provides methods for connecting to the TTS Realtime API, sending text data, retrieving audio output, and managing WebSocket connections.
    */
   public class TTSRealtimeClient {
   
       public enum SessionMode {
           SERVER_COMMIT("server_commit"),
           COMMIT("commit");
           private final String value;
           SessionMode(String value) { this.value = value; }
           public String getValue() { return value; }
       }
   
       /**
        * Audio callback interface
        */
       public interface AudioCallback {
           void onAudio(byte[] audioData);
       }
   
       private final String baseUrl;
       private final String apiKey;
       private final String voice;
       private final SessionMode mode;
       private final String languageType;
       private final AudioCallback audioCallback;
       private final Gson gson = new Gson();
   
       private WebSocketClient ws;
       private CountDownLatch responseDoneLatch;
       private CountDownLatch sessionFinishedLatch;
   
       public TTSRealtimeClient(String baseUrl, String apiKey, String voice,
                                SessionMode mode, AudioCallback audioCallback,
                                String languageType) {
           this.baseUrl = baseUrl;
           this.apiKey = apiKey;
           this.voice = voice;
           this.mode = mode;
           this.audioCallback = audioCallback;
           this.languageType = languageType;
       }
   
       public TTSRealtimeClient(String baseUrl, String apiKey, String voice,
                                SessionMode mode, AudioCallback audioCallback) {
           this(baseUrl, apiKey, voice, mode, audioCallback, "Auto");
       }
   
       /**
        * Establish a WebSocket connection to the TTS Realtime API.
        */
       public void connect() throws Exception {
           Map<String, String> headers = new HashMap<>();
           headers.put("Authorization", "Bearer " + apiKey);
   
           responseDoneLatch = new CountDownLatch(0);
           sessionFinishedLatch = new CountDownLatch(1);
   
           ws = new WebSocketClient(new URI(baseUrl), headers) {
               @Override
               public void onOpen(ServerHandshake handshake) {
                   System.out.println("WebSocket connection established");
                   // Send default session configuration
                   JsonObject session = new JsonObject();
                   session.addProperty("mode", mode.getValue());
                   session.addProperty("voice", TTSRealtimeClient.this.voice);
                   // To use the instruction control feature, uncomment the lines below and replace the model with qwen3-tts-instruct-flash-realtime
                   // session.addProperty("instructions", "Speak quickly with a noticeable rising intonation, suitable for introducing fashion products.");
                   // session.addProperty("optimize_instructions", true);
                   session.addProperty("language_type", languageType);
                   session.addProperty("response_format", "pcm");
                   session.addProperty("sample_rate", 24000);
                   updateSession(session);
               }
   
               @Override
               public void onMessage(String message) {
                   JsonObject event = gson.fromJson(message, JsonObject.class);
                   String eventType = event.has("type") ? event.get("type").getAsString() : "";
   
                   if (!"response.audio.delta".equals(eventType)) {
                       System.out.println("Received event: " + eventType);
                   }
   
                   switch (eventType) {
                       case "error":
                           System.err.println("Error: " + event.get("error"));
                           break;
                       case "session.created":
                           System.out.println("Session created, ID: " +
                               event.getAsJsonObject("session").get("id").getAsString());
                           break;
                       case "session.updated":
                           System.out.println("Session updated, ID: " +
                               event.getAsJsonObject("session").get("id").getAsString());
                           break;
                       case "input_text_buffer.committed":
                           System.out.println("Text buffer committed, item ID: " + event.get("item_id"));
                           break;
                       case "input_text_buffer.cleared":
                           System.out.println("Text buffer cleared");
                           break;
                       case "response.created":
                           System.out.println("Response created, ID: " +
                               event.getAsJsonObject("response").get("id").getAsString());
                           responseDoneLatch = new CountDownLatch(1);
                           break;
                       case "response.output_item.added":
                           System.out.println("Output item added, ID: " +
                               event.getAsJsonObject("item").get("id").getAsString());
                           break;
                       case "response.audio.delta":
                           if (audioCallback != null) {
                               byte[] audioBytes = Base64.getDecoder().decode(
                                   event.get("delta").getAsString());
                               audioCallback.onAudio(audioBytes);
                           }
                           break;
                       case "response.audio.done":
                           System.out.println("Audio generation completed");
                           break;
                       case "response.done":
                           System.out.println("Response completed");
                           responseDoneLatch.countDown();
                           break;
                       case "session.finished":
                           System.out.println("Session finished");
                           sessionFinishedLatch.countDown();
                           break;
                   }
               }
   
               @Override
               public void onClose(int code, String reason, boolean remote) {
                   System.out.println("Connection closed: " + reason);
               }
   
               @Override
               public void onError(Exception ex) {
                   System.err.println("WebSocket error: " + ex.getMessage());
               }
           };
           ws.connectBlocking();
       }
   
       /**
        * Send an event to the server.
        */
       public void sendEvent(JsonObject event) {
           String eventId = "event_" + System.currentTimeMillis();
           event.addProperty("event_id", eventId);
           System.out.println("Sending event: type=" + event.get("type").getAsString()
               + ", event_id=" + eventId);
           ws.send(gson.toJson(event));
       }
   
       /**
        * Update the session configuration.
        */
       public void updateSession(JsonObject config) {
           JsonObject event = new JsonObject();
           event.addProperty("type", "session.update");
           event.add("session", config);
           System.out.println("Updating session configuration: " + event);
           sendEvent(event);
       }
   
       /**
        * Send text data to the API.
        */
       public void appendText(String text) {
           JsonObject event = new JsonObject();
           event.addProperty("type", "input_text_buffer.append");
           event.addProperty("text", text);
           sendEvent(event);
       }
   
       /**
        * Commit the text buffer to trigger processing.
        */
       public void commitTextBuffer() {
           JsonObject event = new JsonObject();
           event.addProperty("type", "input_text_buffer.commit");
           sendEvent(event);
       }
   
       /**
        * Clear the text buffer.
        */
       public void clearTextBuffer() {
           JsonObject event = new JsonObject();
           event.addProperty("type", "input_text_buffer.clear");
           sendEvent(event);
       }
   
       /**
        * End the session.
        */
       public void finishSession() {
           JsonObject event = new JsonObject();
           event.addProperty("type", "session.finish");
           sendEvent(event);
       }
   
       /**
        * Wait for the response.done event.
        */
       public void waitForResponseDone() throws InterruptedException {
           responseDoneLatch.await();
       }
   
       /**
        * Wait for the session.finished event.
        */
       public void waitForSessionFinished() throws InterruptedException {
           sessionFinishedLatch.await();
       }
   
       /**
        * Close the WebSocket connection.
        */
       public void close() {
           if (ws != null) {
               ws.close();
           }
       }
   }
   

2. Selecione um modo de síntese de fala

   A Realtime API oferece suporte a dois modos:

   • Modo server_commit

     O servidor gerencia automaticamente a segmentação de texto e o tempo de síntese. O cliente apenas envia o texto. Ideal para cenários de baixa latência, como navegação GPS.

   • Modo commit

     O cliente adiciona texto a um buffer e aciona explicitamente a síntese. Recomendado para casos que exigem controle preciso da segmentação de frases, como transmissões de notícias.

   Modo server_commit

   Python

   No mesmo diretório do arquivo tts_realtime_client.py, crie outro arquivo Python chamado server_commit.py e copie o código abaixo para ele:

   import os
   import asyncio
   import logging
   import wave
   from tts_realtime_client import TTSRealtimeClient, SessionMode
   import pyaudio
   
   # QwenTTS service configuration
   # To use the instruction control feature, replace the model with qwen3-tts-instruct-flash-realtime and uncomment instructions in tts_realtime_client.py
   # The following is the configuration for the Singapore region.
   URL = "wss://dashscope-intl.aliyuncs.com/api-ws/v1/realtime?model=qwen3-tts-flash-realtime"
   # The API Key differs between the Singapore and Beijing regions. Get your API Key: https://www.alibabacloud.com/help/zh/model-studio/get-api-key
   # If you have not configured the environment variable, replace the next line with your Chinese Model Studio API Key: API_KEY="sk-xxx"
   API_KEY = os.getenv("DASHSCOPE_API_KEY")
   
   if not API_KEY:
       raise ValueError("Please set DASHSCOPE_API_KEY environment variable")
   
   # Collect audio data
   _audio_chunks = []
   # Real-time playback related
   _AUDIO_SAMPLE_RATE = 24000
   _audio_pyaudio = pyaudio.PyAudio()
   _audio_stream = None  # Will be opened at runtime
   
   def _audio_callback(audio_bytes: bytes):
       """TTSRealtimeClient audio callback: real-time playback and caching"""
       global _audio_stream
       if _audio_stream is not None:
           try:
               _audio_stream.write(audio_bytes)
           except Exception as exc:
               logging.error(f"PyAudio playback error: {exc}")
       _audio_chunks.append(audio_bytes)
       logging.info(f"Received audio chunk: {len(audio_bytes)} bytes")
   
   def _save_audio_to_file(filename: str = "output.wav", sample_rate: int = 24000) -> bool:
       """Save collected audio data as a WAV file"""
       if not _audio_chunks:
           logging.warning("No audio data to save")
           return False
   
       try:
           audio_data = b"".join(_audio_chunks)
           with wave.open(filename, 'wb') as wav_file:
               wav_file.setnchannels(1)  # Mono
               wav_file.setsampwidth(2)  # 16-bit
               wav_file.setframerate(sample_rate)
               wav_file.writeframes(audio_data)
           logging.info(f"Audio saved to: {filename}")
           return True
       except Exception as exc:
           logging.error(f"Failed to save audio: {exc}")
           return False
   
   async def _produce_text(client: TTSRealtimeClient):
       """Send text fragments to the server"""
       text_fragments = [
           "Alibaba Cloud's large language model platform, Model Studio, is an all-in-one platform for developing and building large language model applications.",
           "Both developers and business users can deeply participate in the design and development of large language model applications.",
           "You can develop a large language model application in five minutes using a simple interface,",
           "or train a dedicated model in a few hours, allowing you to focus more energy on application innovation.",
       ]
   
       logging.info("Sending text fragments…")
       for text in text_fragments:
           logging.info(f"Sending fragment: {text}")
           await client.append_text(text)
           await asyncio.sleep(0.1)  # Brief delay between fragments
   
       # Wait for the server to finish internal processing before ending the session
       await asyncio.sleep(1.0)
       await client.finish_session()
   
   async def _run_demo():
       """Run the complete demo"""
       global _audio_stream
       # Open PyAudio output stream
       _audio_stream = _audio_pyaudio.open(
           format=pyaudio.paInt16,
           channels=1,
           rate=_AUDIO_SAMPLE_RATE,
           output=True,
           frames_per_buffer=1024
       )
   
       client = TTSRealtimeClient(
           base_url=URL,
           api_key=API_KEY,
           voice="Cherry",
           mode=SessionMode.SERVER_COMMIT,
           audio_callback=_audio_callback
       )
   
       # Establish connection
       await client.connect()
   
       # Run message handling and text sending in parallel
       consumer_task = asyncio.create_task(client.handle_messages())
       producer_task = asyncio.create_task(_produce_text(client))
   
       await producer_task  # Wait for text sending to complete
   
       # Wait for response.done
       await client.wait_for_response_done()
   
       # Close connection and cancel consumer task
       await client.close()
       consumer_task.cancel()
   
       # Close audio stream
       if _audio_stream is not None:
           _audio_stream.stop_stream()
           _audio_stream.close()
       _audio_pyaudio.terminate()
   
       # Save audio data
       os.makedirs("outputs", exist_ok=True)
       _save_audio_to_file(os.path.join("outputs", "qwen_tts_output.wav"))
   
   def main():
       """Synchronous entry point"""
       logging.basicConfig(
           level=logging.INFO,
           format='%(asctime)s [%(levelname)s] %(message)s',
           datefmt='%Y-%m-%d %H:%M:%S'
       )
       logging.info("Starting QwenTTS Realtime Client demo…")
       asyncio.run(_run_demo())
   
   if __name__ == "__main__":
       main()
   

   Execute o arquivo server_commit.py para ouvir em tempo real o áudio gerado pela Realtime API.

   Java

   No mesmo diretório do arquivo TTSRealtimeClient.java, crie outro arquivo Java chamado ServerCommit.java e copie o código abaixo para ele:

   import javax.sound.sampled.*;
   import java.io.*;
   import java.util.ArrayList;
   import java.util.List;
   import java.util.concurrent.ConcurrentLinkedQueue;
   import java.util.concurrent.atomic.AtomicBoolean;
   
   public class ServerCommit {
       // The following is the configuration for the Singapore region.
       private static final String URL = "wss://dashscope-intl.aliyuncs.com/api-ws/v1/realtime?model=qwen3-tts-flash-realtime";
       // The API Key differs between the Singapore and Beijing regions. Get your API Key: https://www.alibabacloud.com/help/zh/model-studio/get-api-key
       // If you have not configured the environment variable, replace the next line with your Chinese Model Studio API Key: private static final String API_KEY = "sk-xxx";
       private static final String API_KEY = System.getenv("DASHSCOPE_API_KEY");
       private static final int SAMPLE_RATE = 24000;
   
       // Audio data cache
       private static final List<byte[]> audioChunks = new ArrayList<>();
       // Real-time playback queue
       private static final ConcurrentLinkedQueue<byte[]> playbackQueue = new ConcurrentLinkedQueue<>();
       private static final AtomicBoolean playing = new AtomicBoolean(true);
   
       public static void main(String[] args) throws Exception {
           if (API_KEY == null || API_KEY.isEmpty()) {
               throw new IllegalStateException("Please set the DASHSCOPE_API_KEY environment variable");
           }
   
           // Initialize audio playback
           AudioFormat format = new AudioFormat(SAMPLE_RATE, 16, 1, true, false);
           DataLine.Info info = new DataLine.Info(SourceDataLine.class, format);
           SourceDataLine audioLine = (SourceDataLine) AudioSystem.getLine(info);
           audioLine.open(format);
           audioLine.start();
   
           // Start the playback thread
           Thread playerThread = new Thread(() -> {
               while (playing.get() || !playbackQueue.isEmpty()) {
                   byte[] chunk = playbackQueue.poll();
                   if (chunk != null) {
                       audioLine.write(chunk, 0, chunk.length);
                   } else {
                       try { Thread.sleep(10); } catch (InterruptedException ignored) {}
                   }
               }
           });
           playerThread.start();
   
           // Create the TTS client
           // To use the instruction control feature, replace the model with qwen3-tts-instruct-flash-realtime and uncomment the instructions in TTSRealtimeClient.java
           TTSRealtimeClient client = new TTSRealtimeClient(
               URL, API_KEY, "Cherry",
               TTSRealtimeClient.SessionMode.SERVER_COMMIT,
               audioData -> {
                   playbackQueue.add(audioData);
                   audioChunks.add(audioData);
                   System.out.println("Received audio data: " + audioData.length + " bytes");
               }
           );
   
           client.connect();
   
           // Send text fragments
           String[] textFragments = {
               "Alibaba Cloud's large language model platform, Model Studio, is an all-in-one platform for developing and building large language model applications.",
               "Both developers and business users can deeply participate in the design and development of large language model applications.",
               "You can develop a large language model application in five minutes using a simple interface,",
               "or train a dedicated model in a few hours, allowing you to focus more energy on application innovation."
           };
   
           System.out.println("Starting to send text...");
           for (String text : textFragments) {
               System.out.println("Sending fragment: " + text);
               client.appendText(text);
               Thread.sleep(100);
           }
   
           Thread.sleep(1000);
           client.finishSession();
   
           // Wait for the response to complete
           client.waitForResponseDone();
           client.waitForSessionFinished();
           client.close();
   
           // Wait for playback to complete
           playing.set(false);
           playerThread.join();
           audioLine.drain();
           audioLine.close();
   
           // Save the audio file
           saveWav("output.wav");
           System.out.println("Done");
       }
   
       private static void saveWav(String filename) throws IOException {
           if (audioChunks.isEmpty()) {
               System.out.println("No audio data to save");
               return;
           }
           ByteArrayOutputStream bos = new ByteArrayOutputStream();
           for (byte[] chunk : audioChunks) {
               bos.write(chunk);
           }
           byte[] allAudio = bos.toByteArray();
           AudioFormat format = new AudioFormat(SAMPLE_RATE, 16, 1, true, false);
           AudioInputStream ais = new AudioInputStream(
               new ByteArrayInputStream(allAudio), format, allAudio.length / 2);
           new File("outputs").mkdirs();
           AudioSystem.write(ais, AudioFileFormat.Type.WAVE,
               new File("outputs/" + filename));
           System.out.println("Audio saved to: outputs/" + filename);
       }
   }
   

   Compile e execute o arquivo ServerCommit.java para ouvir em tempo real o áudio gerado pela Realtime API.

   Modo commit

   Python

   No mesmo diretório do arquivo tts_realtime_client.py, crie outro arquivo Python chamado commit.py e copie o código abaixo para ele:

   import os
   import asyncio
   import logging
   import wave
   from tts_realtime_client import TTSRealtimeClient, SessionMode
   import pyaudio
   
   # QwenTTS service configuration
   # To use the instruction control feature, replace the model with qwen3-tts-instruct-flash-realtime and uncomment instructions in tts_realtime_client.py
   # The following is the configuration for the Singapore region.
   URL = "wss://dashscope-intl.aliyuncs.com/api-ws/v1/realtime?model=qwen3-tts-flash-realtime"
   # The API Key differs between the Singapore and Beijing regions. Get your API Key: https://www.alibabacloud.com/help/zh/model-studio/get-api-key
   # If you have not configured the environment variable, replace the next line with your Chinese Model Studio API Key: API_KEY="sk-xxx"
   API_KEY = os.getenv("DASHSCOPE_API_KEY")
   
   if not API_KEY:
       raise ValueError("Please set DASHSCOPE_API_KEY environment variable")
   
   # Collect audio data
   _audio_chunks = []
   _AUDIO_SAMPLE_RATE = 24000
   _audio_pyaudio = pyaudio.PyAudio()
   _audio_stream = None
   
   def _audio_callback(audio_bytes: bytes):
       """TTSRealtimeClient audio callback: real-time playback and caching"""
       global _audio_stream
       if _audio_stream is not None:
           try:
               _audio_stream.write(audio_bytes)
           except Exception as exc:
               logging.error(f"PyAudio playback error: {exc}")
       _audio_chunks.append(audio_bytes)
       logging.info(f"Received audio chunk: {len(audio_bytes)} bytes")
   
   def _save_audio_to_file(filename: str = "output.wav", sample_rate: int = 24000) -> bool:
       """Save collected audio data as a WAV file"""
       if not _audio_chunks:
           logging.warning("No audio data to save")
           return False
   
       try:
           audio_data = b"".join(_audio_chunks)
           with wave.open(filename, 'wb') as wav_file:
               wav_file.setnchannels(1)  # Mono
               wav_file.setsampwidth(2)  # 16-bit
               wav_file.setframerate(sample_rate)
               wav_file.writeframes(audio_data)
           logging.info(f"Audio saved to: {filename}")
           return True
       except Exception as exc:
           logging.error(f"Failed to save audio: {exc}")
           return False
   
   async def _user_input_loop(client: TTSRealtimeClient):
       """Continuously get user input and send text. When user enters empty text, send a commit event and end the current session"""
       print("Enter text (press Enter directly to send a commit event and end the current session, press Ctrl+C or Ctrl+D to exit the program):")
   
       while True:
           try:
               user_text = input("> ")
               if not user_text:  # User input is empty
                   # Empty input is treated as end of a conversation: commit buffer -> end session -> break loop
                   logging.info("Empty input, sending commit event and ending current session")
                   await client.commit_text_buffer()
                   # Wait briefly for the server to process the commit, preventing premature session end that could lose audio
                   await asyncio.sleep(0.3)
                   await client.finish_session()
                   break  # Exit user input loop directly, no need to press Enter again
               else:
                   logging.info(f"Sending text: {user_text}")
                   await client.append_text(user_text)
   
           except EOFError:  # User pressed Ctrl+D
               break
           except KeyboardInterrupt:  # User pressed Ctrl+C
               break
   
       # End session
       logging.info("Ending session...")
   async def _run_demo():
       """Run the complete demo"""
       global _audio_stream
       # Open PyAudio output stream
       _audio_stream = _audio_pyaudio.open(
           format=pyaudio.paInt16,
           channels=1,
           rate=_AUDIO_SAMPLE_RATE,
           output=True,
           frames_per_buffer=1024
       )
   
       client = TTSRealtimeClient(
           base_url=URL,
           api_key=API_KEY,
           voice="Cherry",
           mode=SessionMode.COMMIT,  # Changed to COMMIT mode
           audio_callback=_audio_callback
       )
   
       # Establish connection
       await client.connect()
   
       # Run message handling and user input in parallel
       consumer_task = asyncio.create_task(client.handle_messages())
       producer_task = asyncio.create_task(_user_input_loop(client))
   
       await producer_task  # Wait for user input to complete
   
       # Wait for response.done
       await client.wait_for_response_done()
   
       # Close connection and cancel consumer task
       await client.close()
       consumer_task.cancel()
   
       # Close audio stream
       if _audio_stream is not None:
           _audio_stream.stop_stream()
           _audio_stream.close()
       _audio_pyaudio.terminate()
   
       # Save audio data
       os.makedirs("outputs", exist_ok=True)
       _save_audio_to_file(os.path.join("outputs", "qwen_tts_output.wav"))
   
   def main():
       logging.basicConfig(
           level=logging.INFO,
           format='%(asctime)s [%(levelname)s] %(message)s',
           datefmt='%Y-%m-%d %H:%M:%S'
       )
       logging.info("Starting QwenTTS Realtime Client demo…")
       asyncio.run(_run_demo())
   
   if __name__ == "__main__":
       main()
   

   Execute o arquivo commit.py. É possível inserir texto para síntese quantas vezes desejar. Pressione Enter sem digitar nenhum texto para ouvir, pelo alto-falante, o áudio retornado pela Realtime API.

   Java

   No mesmo diretório do arquivo TTSRealtimeClient.java, crie outro arquivo Java chamado Commit.java e copie o código abaixo para ele:

   import javax.sound.sampled.*;
   import java.io.*;
   import java.util.ArrayList;
   import java.util.List;
   import java.util.Scanner;
   import java.util.concurrent.ConcurrentLinkedQueue;
   import java.util.concurrent.atomic.AtomicBoolean;
   
   public class Commit {
       // The following is the configuration for the Singapore region.
       private static final String URL = "wss://dashscope-intl.aliyuncs.com/api-ws/v1/realtime?model=qwen3-tts-flash-realtime";
       // The API Key differs between the Singapore and Beijing regions. Get your API Key: https://www.alibabacloud.com/help/zh/model-studio/get-api-key
       // If you have not configured the environment variable, replace the next line with your Chinese Model Studio API Key: private static final String API_KEY = "sk-xxx";
       private static final String API_KEY = System.getenv("DASHSCOPE_API_KEY");
       private static final int SAMPLE_RATE = 24000;
   
       private static final List<byte[]> audioChunks = new ArrayList<>();
       private static final ConcurrentLinkedQueue<byte[]> playbackQueue = new ConcurrentLinkedQueue<>();
       private static final AtomicBoolean playing = new AtomicBoolean(true);
   
       public static void main(String[] args) throws Exception {
           if (API_KEY == null || API_KEY.isEmpty()) {
               throw new IllegalStateException("Please set the DASHSCOPE_API_KEY environment variable");
           }
   
           // Initialize audio playback
           AudioFormat format = new AudioFormat(SAMPLE_RATE, 16, 1, true, false);
           DataLine.Info info = new DataLine.Info(SourceDataLine.class, format);
           SourceDataLine audioLine = (SourceDataLine) AudioSystem.getLine(info);
           audioLine.open(format);
           audioLine.start();
   
           // Start the playback thread
           Thread playerThread = new Thread(() -> {
               while (playing.get() || !playbackQueue.isEmpty()) {
                   byte[] chunk = playbackQueue.poll();
                   if (chunk != null) {
                       audioLine.write(chunk, 0, chunk.length);
                   } else {
                       try { Thread.sleep(10); } catch (InterruptedException ignored) {}
                   }
               }
           });
           playerThread.start();
   
           // Create a TTS client (commit mode)
           // To use instruction control, replace the model with qwen3-tts-instruct-flash-realtime and uncomment the instructions in TTSRealtimeClient.java
           TTSRealtimeClient client = new TTSRealtimeClient(
               URL, API_KEY, "Cherry",
               TTSRealtimeClient.SessionMode.COMMIT,
               audioData -> {
                   playbackQueue.add(audioData);
                   audioChunks.add(audioData);
                   System.out.println("Received audio data: " + audioData.length + " bytes");
               }
           );
   
           client.connect();
   
           // Interactive input
           System.out.println("Enter text (press Enter directly to send a commit event and end the session, press Ctrl+D to exit the program):");
           Scanner scanner = new Scanner(System.in);
           while (true) {
               System.out.print("> ");
               if (!scanner.hasNextLine()) {
                   client.finishSession();
                   break;
               }
               String userText = scanner.nextLine();
               if (userText.isEmpty()) {
                   // Empty input: commit the buffer and end the session
                   System.out.println("Empty input, sending commit event and ending the session");
                   client.commitTextBuffer();
                   Thread.sleep(300);
                   client.finishSession();
                   break;
               } else {
                   System.out.println("Sending text: " + userText);
                   client.appendText(userText);
               }
           }
           scanner.close();
   
           // Wait for the response to complete
           client.waitForResponseDone();
           client.waitForSessionFinished();
           client.close();
   
           // Wait for playback to complete
           playing.set(false);
           playerThread.join();
           audioLine.drain();
           audioLine.close();
   
           // Save the audio file
           saveWav("output.wav");
           System.out.println("Done");
       }
   
       private static void saveWav(String filename) throws IOException {
           if (audioChunks.isEmpty()) {
               System.out.println("No audio data to save");
               return;
           }
           ByteArrayOutputStream bos = new ByteArrayOutputStream();
           for (byte[] chunk : audioChunks) {
               bos.write(chunk);
           }
           byte[] allAudio = bos.toByteArray();
           AudioFormat format = new AudioFormat(SAMPLE_RATE, 16, 1, true, false);
           AudioInputStream ais = new AudioInputStream(
               new ByteArrayInputStream(allAudio), format, allAudio.length / 2);
           new File("outputs").mkdirs();
           AudioSystem.write(ais, AudioFileFormat.Type.WAVE,
               new File("outputs/" + filename));
           System.out.println("Audio saved to: outputs/" + filename);
       }
   }
   

   Compile e execute o arquivo Commit.java. É possível inserir texto para síntese quantas vezes desejar. Pressione Enter sem digitar nenhum texto para ouvir, pelo alto-falante, o áudio retornado pela Realtime API.

Qwen-Audio-TTS/CosyVoice

Qwen-Audio-TTS e Qwen-Audio-TTS/CosyVoice compartilham a mesma interface do SDK. Os exemplos abaixo também se aplicam aos modelos Qwen-Audio-TTS — basta substituir os parâmetros model e voice.

Pré-requisitos

• Obtain an API key

• DashScope SDK instalado e compatível com os requisitos de versão. Recomendamos installing the latest version:

  • Python SDK: versão >= 1.25.2
  • Java SDK: versão >= 2.16.6

from dashscope.audio.tts_v2 import SpeechSynthesizerObjectPool

synthesizer_object_pool = SpeechSynthesizerObjectPool(max_size=20)
import dashscope
# The following is the configuration for the China (Beijing) region. Replace "{WorkspaceId}" with your actual workspace ID. The configuration varies by region.
dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"

speech_synthesizer = connectionPool.borrow_synthesizer(
    model='cosyvoice-v3-flash',
    voice='longanyang',
    seed=12382,
    callback=synthesizer_callback
)

connectionPool.return_synthesizer(speech_synthesizer)

# !/usr/bin/env python3
# Copyright (C) Alibaba Group. All Rights Reserved.
# MIT License (https://opensource.org/licenses/MIT)

import os
import time
import threading

import dashscope
from dashscope.audio.tts_v2 import *

USE_CONNECTION_POOL = True
text_to_synthesize = [
    'Sentence 1: Welcome to Alibaba speech synthesis service.',
    'Sentence 2: Welcome to Alibaba speech synthesis service.',
    'Sentence 3: Welcome to Alibaba speech synthesis service.',
]
connectionPool = None

def init_dashscope_api_key():
    '''
    Set your DashScope API-key. More information:
    https://github.com/aliyun/alibabacloud-bailian-speech-demo/blob/master/PREREQUISITES.md
    '''
    # The API Key differs between the Singapore and Beijing regions. Get your API Key: https://www.alibabacloud.com/help/zh/model-studio/get-api-key
    if 'DASHSCOPE_API_KEY' in os.environ:
        dashscope.api_key = os.environ[
            'DASHSCOPE_API_KEY']  # load API-key from environment variable DASHSCOPE_API_KEY
    else:
        dashscope.api_key = '<your-dashscope-api-key>'  # set API-key manually

def synthesis_text_to_speech_and_play_by_streaming_mode(text, task_id):
    global USE_CONNECTION_POOL, connectionPool
    '''
    Synthesize speech with given text by streaming mode, async call and play the synthesized audio in real-time.
    for more information, please refer to https://www.alibabacloud.com/help/document_detail/2712523.html
    '''

    complete_event = threading.Event()

    # Define a callback to handle the result

    class Callback(ResultCallback):
        def on_open(self):
            # when using object pool, on_open will be called after task start
            self.file = open(f'result_{task_id}.mp3', 'wb')
            print(f'[task_{task_id}] start')

        def on_complete(self):
            print(f'[task_{task_id}] speech synthesis task complete successfully.')
            complete_event.set()

        def on_error(self, message: str):
            print(f'[task_{task_id}] speech synthesis task failed, {message}')

        def on_close(self):
            # when using object pool, on_open will be called after task finished
            print(f'[task_{task_id}] finished')

        def on_event(self, message):
            # print(f'recv speech synthsis message {message}')
            pass

        def on_data(self, data: bytes) -> None:
            # send to player
            # save audio to file
            self.file.write(data)

    # Call the speech synthesizer callback
    synthesizer_callback = Callback()

    # Initialize the speech synthesizer
    # you can customize the synthesis parameters, like voice, format, sample_rate or other parameters
    if USE_CONNECTION_POOL:
        speech_synthesizer = connectionPool.borrow_synthesizer(
            model='cosyvoice-v3-flash',
            voice='longanyang',
            seed=12382,
            callback=synthesizer_callback
        )
    else:
        speech_synthesizer = SpeechSynthesizer(model='cosyvoice-v3-flash',
                                               voice='longanyang',
                                               seed=12382,
                                               callback=synthesizer_callback)
    try:
        speech_synthesizer.call(text)
    except Exception as e:
        print(f'[task_{task_id}] speech synthesis task failed, {e}')
        if USE_CONNECTION_POOL:
            # close the synthesizer connection manually if task failed when using connection pool.
            speech_synthesizer.close()
        return

    print('[task_{}] Synthesized text: {}'.format(task_id, text))
    complete_event.wait()
    print('[task_{}][Metric] requestId: {}, first package delay ms: {}'.format(
        task_id,
        speech_synthesizer.get_last_request_id(),
        speech_synthesizer.get_first_package_delay()))
    if USE_CONNECTION_POOL:
        connectionPool.return_synthesizer(speech_synthesizer)

# main function
if __name__ == '__main__':
    # You must set dashscope.api_key and base_websocket_api_url before creating SpeechSynthesizerObjectPool.
    # The pool establishes WebSocket connections using the current global dashscope.api_key at initialization time.
    # Modifying dashscope.api_key after pool creation will not affect existing connections in the pool.
    # The following is the configuration for the Singapore region. Replace "{WorkspaceId}" with your actual workspace ID. The configuration varies by region.
    dashscope.base_websocket_api_url='wss://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api-ws/v1/inference'
    init_dashscope_api_key()

    if USE_CONNECTION_POOL:
        print('creating connection pool')
        start_time = time.time() * 1000
        connectionPool = SpeechSynthesizerObjectPool(max_size=3)
        end_time = time.time() * 1000
        print('connection pool created, cost: {} ms'.format(end_time - start_time))

    task_thread_list = []
    for task_id in range(3):
        thread = threading.Thread(
            target=synthesis_text_to_speech_and_play_by_streaming_mode,
            args=(text_to_synthesize[task_id], task_id))
        task_thread_list.append(thread)

    for task_thread in task_thread_list:
        task_thread.start()

    for task_thread in task_thread_list:
        task_thread.join()

    if USE_CONNECTION_POOL:
        connectionPool.shutdown()

class CosyvoiceObjectPool {
    // ... Other code is omitted here. For the complete example, see the full code.
    public static GenericObjectPool<SpeechSynthesizer> getInstance() {
        lock.lock();
        if (synthesizerPool == null) {
            // You can set the object pool size here, or set it in the environment variable COSYVOICE_OBJECTPOOL_SIZE.
            // It is recommended to set it to 1.5 to 2 times the maximum concurrent connections of the server.
            int objectPoolSize = getObjectivePoolSize();
            SpeechSynthesizerObjectFactory speechSynthesizerObjectFactory =
                    new SpeechSynthesizerObjectFactory();
            GenericObjectPoolConfig<SpeechSynthesizer> config =
                    new GenericObjectPoolConfig<>();
            config.setMaxTotal(objectPoolSize);
            config.setMaxIdle(objectPoolSize);
            config.setMinIdle(objectPoolSize);
            synthesizerPool =
                    new GenericObjectPool<>(speechSynthesizerObjectFactory, config);
        }
        lock.unlock();
        return synthesizerPool;
    }
}

synthesizer = CosyvoiceObjectPool.getInstance().borrowObject();

CosyvoiceObjectPool.getInstance().returnObject(synthesizer);

import com.alibaba.dashscope.audio.tts.SpeechSynthesisResult;
import com.alibaba.dashscope.audio.ttsv2.SpeechSynthesisAudioFormat;
import com.alibaba.dashscope.audio.ttsv2.SpeechSynthesisParam;
import com.alibaba.dashscope.audio.ttsv2.SpeechSynthesizer;
import com.alibaba.dashscope.common.ResultCallback;
import com.alibaba.dashscope.exception.NoApiKeyException;
import com.alibaba.dashscope.utils.Constants;
import lombok.extern.slf4j.Slf4j;
import org.apache.commons.pool2.BasePooledObjectFactory;
import org.apache.commons.pool2.PooledObject;
import org.apache.commons.pool2.impl.DefaultPooledObject;
import org.apache.commons.pool2.impl.GenericObjectPool;
import org.apache.commons.pool2.impl.GenericObjectPoolConfig;

import java.time.LocalDateTime;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.Lock;

/**
 * You need to include the org.apache.commons.pool2 and DashScope packages in your project.
 *
 * DashScope SDK 2.16.6 and later versions are optimized for high-concurrency scenarios.
 * Versions earlier than DashScope SDK 2.16.6 are not recommended for high-concurrency use.
 *
 *
 * Before making high-concurrency calls to the TTS service,
 * please configure the connection pool parameters through the following environment variables.
 *
 * DASHSCOPE_MAXIMUM_ASYNC_REQUESTS
 * DASHSCOPE_MAXIMUM_ASYNC_REQUESTS_PER_HOST
 * DASHSCOPE_CONNECTION_POOL_SIZE
 *
 */

class SpeechSynthesizerObjectFactory
        extends BasePooledObjectFactory<SpeechSynthesizer> {
    public SpeechSynthesizerObjectFactory() {
        super();
    }
    @Override
    public SpeechSynthesizer create() throws Exception {
        return new SpeechSynthesizer();
    }

    @Override
    public PooledObject<SpeechSynthesizer> wrap(SpeechSynthesizer obj) {
        return new DefaultPooledObject<>(obj);
    }
}

class CosyvoiceObjectPool {
    public static GenericObjectPool<SpeechSynthesizer> synthesizerPool;
    public static String COSYVOICE_OBJECTPOOL_SIZE_ENV = "COSYVOICE_OBJECTPOOL_SIZE";
    public static int DEFAULT_OBJECT_POOL_SIZE = 500;
    private static Lock lock = new java.util.concurrent.locks.ReentrantLock();
    public static int getObjectivePoolSize() {
        try {
            Integer n = Integer.parseInt(System.getenv(COSYVOICE_OBJECTPOOL_SIZE_ENV));
            System.out.println("Using Object Pool Size In Env: "+ n);
            return n;
        } catch (NumberFormatException e) {
            System.out.println("Using Default Object Pool Size: "+ DEFAULT_OBJECT_POOL_SIZE);
            return DEFAULT_OBJECT_POOL_SIZE;
        }
    }
    public static GenericObjectPool<SpeechSynthesizer> getInstance() {
        lock.lock();
        if (synthesizerPool == null) {
            // You can set the object pool size here or in the COSYVOICE_OBJECTPOOL_SIZE environment variable.
            // It is recommended to set it to 1.5 to 2 times your server's maximum concurrent connections.
            int objectPoolSize = getObjectivePoolSize();
            SpeechSynthesizerObjectFactory speechSynthesizerObjectFactory =
                    new SpeechSynthesizerObjectFactory();
            GenericObjectPoolConfig<SpeechSynthesizer> config =
                    new GenericObjectPoolConfig<>();
            config.setMaxTotal(objectPoolSize);
            config.setMaxIdle(objectPoolSize);
            config.setMinIdle(objectPoolSize);
            synthesizerPool =
                    new GenericObjectPool<>(speechSynthesizerObjectFactory, config);
        }
        lock.unlock();
        return synthesizerPool;
    }
}

class SynthesizeTaskWithCallback implements Runnable {
    String[] textArray;
    String requestId;
    long timeCost;
    public SynthesizeTaskWithCallback(String[] textArray) {
        this.textArray = textArray;
    }
    @Override
    public void run() {
        SpeechSynthesizer synthesizer = null;
        long startTime = System.currentTimeMillis();
        // if recv onError
        final boolean[] hasError = {false};
        try {
            class ReactCallback extends ResultCallback<SpeechSynthesisResult> {
                ReactCallback() {}

                @Override
                public void onEvent(SpeechSynthesisResult message) {
                    if (message.getAudioFrame() != null) {
                        try {
                            byte[] bytesArray = message.getAudioFrame().array();
                            System.out.println("Received audio, audio stream length: " + bytesArray.length);
                        } catch (Exception e) {
                            throw new RuntimeException(e);
                        }
                    }
                }

                @Override
                public void onComplete() {}

                @Override
                public void onError(Exception e) {
                    System.out.println(e.getMessage());
                    e.printStackTrace();
                    hasError[0] = true;
                }
            }

            SpeechSynthesisParam param =
                    SpeechSynthesisParam.builder()
                            .model("cosyvoice-v3-flash")
                            .voice("longanyang")
                            // The API Key differs between the Singapore and Beijing regions. Get your API Key: https://www.alibabacloud.com/help/zh/model-studio/get-api-key
                            // If you have not configured the environment variable, replace the next line with your Chinese Model Studio API Key: .apiKey("sk-xxx")
                            .apiKey(System.getenv("DASHSCOPE_API_KEY"))
                            .format(SpeechSynthesisAudioFormat
                                    .MP3_22050HZ_MONO_256KBPS) // Use PCM or MP3 for streaming synthesis
                            .build();

            try {
                synthesizer = CosyvoiceObjectPool.getInstance().borrowObject();
                // Note: In an object pool scenario, the apiKey passed in multiple calls to updateParamAndCallback must always be the same. The SDK does not update the apiKey for an established connection, so passing a different apiKey will not take effect. See the important notes in the "Perform speech synthesis" step above.
                synthesizer.updateParamAndCallback(param, new ReactCallback());
                for (String text : textArray) {
                    synthesizer.streamingCall(text);
                }
                Thread.sleep(20);
                synthesizer.streamingComplete(60000);
                requestId = synthesizer.getLastRequestId();
            } catch (Exception e) {
                System.out.println("Exception e: " + e.toString());
                hasError[0] = true;
            }
        } catch (Exception e) {
            hasError[0] = true;
            throw new RuntimeException(e);
        }
        if (synthesizer != null) {
            try {
                if (hasError[0] == true) {
                    // If an exception occurs, close the connection and invalidate the object in the pool.
                    synthesizer.getDuplexApi().close(1000, "bye");
                    CosyvoiceObjectPool.getInstance().invalidateObject(synthesizer);
                } else {
                    // If the task completes normally, return the object to the pool.
                    CosyvoiceObjectPool.getInstance().returnObject(synthesizer);
                }
            } catch (Exception e) {
                throw new RuntimeException(e);
            }
            long endTime = System.currentTimeMillis();
            timeCost = endTime - startTime;
            System.out.println("[Thread " + Thread.currentThread() + "] Speech synthesis task completed. Time cost: " + timeCost + " ms, RequestId " + requestId);
        }
    }
}

@Slf4j
public class SynthesizeTextToSpeechWithCallbackConcurrently {
    public static void checkoutEnv(String envName, int defaultSize) {
        if (System.getenv(envName) != null) {
            System.out.println("[ENV CHECK]: " + envName + " "
                    + System.getenv(envName));
        } else {
            System.out.println("[ENV CHECK]: " + envName
                    + " Using Default which is " + defaultSize);
        }
    }

    public static void main(String[] args)
            throws InterruptedException, NoApiKeyException {
        // The following is the configuration for the Singapore region. Replace "{WorkspaceId}" with your actual workspace ID. The configuration varies by region.
        Constants.baseWebsocketApiUrl = "wss://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api-ws/v1/inference";
        // Check for connection pool env
        checkoutEnv("DASHSCOPE_CONNECTION_POOL_SIZE", 32);
        checkoutEnv("DASHSCOPE_MAXIMUM_ASYNC_REQUESTS", 32);
        checkoutEnv("DASHSCOPE_MAXIMUM_ASYNC_REQUESTS_PER_HOST", 32);
        checkoutEnv(CosyvoiceObjectPool.COSYVOICE_OBJECTPOOL_SIZE_ENV, CosyvoiceObjectPool.DEFAULT_OBJECT_POOL_SIZE);

        int runTimes = 3;
        // Create the pool of SpeechSynthesis objects
        ExecutorService executorService = Executors.newFixedThreadPool(runTimes);

        for (int i = 0; i < runTimes; i++) {
            // Record the task submission time
            LocalDateTime submissionTime = LocalDateTime.now();
            executorService.submit(new SynthesizeTaskWithCallback(new String[] {
                    "Before my bed, moonlight shines bright,", "I wonder if it is frost on the ground,", "I raise my eyes to gaze at the bright moon,", "then bow my head, thinking of home."}));
        }

        // Shut down the ExecutorService and wait for all tasks to complete
        executorService.shutdown();
        executorService.awaitTermination(1, TimeUnit.MINUTES);
        System.exit(0);
    }
}

// In the current code, the corresponding content is as follows
// Close the connection
synthesizer.getDuplexApi().close(1000, "bye");
// Invalidate the synthesizer in the object pool when an exception occurs
CosyvoiceObjectPool.getInstance().invalidateObject(synthesizer);