API Documentation

Base URL

All API paths start with /v1, OpenAI-compatible.

Authentication

Send API Key in request header:

Authorization: Bearer sk-wxtts-xxx

Create a key in Dashboard → API Keys. Key prefix: sk-wxtts-

POST /audio/speech

Text to speech, OpenAI-compatible.

curl -X POST https://api.wxtts.com/v1/audio/speech \
  -H "Authorization: Bearer sk-wxtts-xxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-1",
    "input": "你好，欢迎使用王小听。",
    "voice": "xiaoxiao",
    "response_format": "mp3",
    "speed": 1.0,
    "pitch": 1.0
  }' --output speech.mp3

Response Format

Returns JSON by default, audio encoded as base64:

{
  "code": 0,
  "data": {
    "audioBase64": "UklGRiQAAABXQVZF...",
    "format": "mp3",
    "charCount": 12,
    "voice": "xiaoxiao",
    "model": "tts-1",
    "timestamps": [
      { "beginTime": 0, "endTime": 1500, "text": "你好，欢迎使用王小听。" }
    ]
  }
}

Set Accept: audio/mpeg to receive raw binary audio stream. Response headers include X-Char-Count, X-Voice, X-Model.

Other Endpoints

• GET /v1/voices — List voices
• GET /v1/models — List available models
• GET /v1/usage — Query current usage

SSML Support

Use SSML tags in the input field to control pauses, rate/pitch, emphasis, etc. The backend parses tags and stitches segments; plain text without tags is processed as-is.

<speak>
  你好，欢迎使用王小听。
  <break time="500ms"/>
  <prosody rate="0.8" pitch="1.1">这里是慢速高音调片段。</prosody>
  <emphasis>重点强调内容</emphasis>
</speak>

Multi-speaker Dialog

Wrap multiple <speaker voice="voiceId">text</speaker> segments in <dialog> to produce multi-character voiceover. Each speaker uses its own voice, with 300ms auto-pause between turns.

<dialog>
<speaker voice="longanhuan">你好，今天天气真好。</speaker>
<speaker voice="longanyang">是啊，我们去公园散步吧。</speaker>
<speaker voice="longanhuan">好主意，走吧！</speaker>
</dialog>

Streaming TTS

POST /v1/audio/speech/stream splits text by sentence and pushes PCM byte stream (16-bit mono) immediately as each segment completes, achieving low first-byte latency. Content-Type: audio/pcm. Use fetch + ReadableStream on frontend, decode with AudioContext. Stream endpoint outputs PCM only, no mp3/wav.

// 浏览器 fetch + ReadableStream 流式播放
const resp = await fetch("https://api.wxtts.com/v1/audio/speech/stream", {
  method: "POST",
  headers: {
    "Authorization": "Bearer sk-wxtts-xxx",
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
    model: "tts-1",
    input: "长文本将按句切分，每段合成完成立即推送 PCM 流，实现首字低延迟。",
    voice: "longanhuan"
  })
});

const reader = resp.body.getReader();
const audioCtx = new AudioContext({ sampleRate: 22050 });
while (true) {
  const { done, value } = await reader.read();
  if (done) break;
  // value 为 16-bit PCM 字节，用 AudioBufferSourceNode 播放
  const audioBuffer = pcmToAudioBuffer(value, 22050);
  const src = audioCtx.createBufferSource();
  src.buffer = audioBuffer;
  src.connect(audioCtx.destination);
  src.start();
}

Error Codes

Rate Limits

Returns 429 when exceeded.