TTS 文字转语音

请求体为 JSON。成功时响应体是二进制音频流，不是 JSON。

The body is JSON. On success the response is a binary audio stream, not JSON.

成功时返回二进制音频流：

失败时返回 JSON 错误体，结构为 { error: { code, message, type } }。

On success it returns a binary audio stream:

On failure it returns a JSON error body structured as { error: { code, message, type } }.

先用一段代表性文本把六个音色都合成一遍对比，比看文字描述准得多。

Synthesize one representative piece of text with all six voices and compare — far more reliable than reading descriptions.

单次 4096 字符的上限决定了长文必须分段。按句子边界切分比按字数切分自然，能避免句子中间断开。

import re

def split_text(text, limit=3800):
    # 先按句号、问号、叹号切，再按长度贪心合并
    sentences = re.split(r"(?<=[。！？.!?])", text)
    chunks, cur = [], ""
    for s in sentences:
        if len(cur) + len(s) > limit:
            chunks.append(cur)
            cur = s
        else:
            cur += s
    if cur:
        chunks.append(cur)
    return chunks

为什么用 3800 而不是 4096：中文字符在部分实现里按字节计，留 10% 余量更稳妥。

The 4096-character ceiling means long text must be split. Splitting on sentence boundaries is more natural than splitting by character count and avoids chopping mid-sentence.

import re

def split_text(text, limit=3800):
    # split on sentence enders first, then greedily merge by length
    sentences = re.split(r"(?<=[。！？.!?])", text)
    chunks, cur = [], ""
    for s in sentences:
        if len(cur) + len(s) > limit:
            chunks.append(cur)
            cur = s
        else:
            cur += s
    if cur:
        chunks.append(cur)
    return chunks

Why 3800 rather than 4096: some implementations count Chinese characters by bytes, so a 10% margin is safer.

更多状态码与重试建议见错误码与限流，常见报错的排查步骤见常见问题。

For more status codes and retry guidance see Error codes and rate limits; for troubleshooting steps on common errors see the FAQ.

以下语音合成模型均可通过本接口调用，实际可用范围以你的账号分组为准：

Every speech synthesis model below can be called through this endpoint; what your account group actually allows takes precedence:

基础合成

传文本、音色与输出格式，直接拿回音频二进制。默认 24kHz 采样率。

音质优先

tts-1-hd 音质更好但延迟更高，适合预生成的有声内容，不适合实时对话。

调整语速

speed 取 0.25 到 4.0，1.0 是原速。教学类内容常降到 0.9 左右，方便跟读。

流式合成

需要边生成边播放时用流式格式，客户端收到首块就能开始播，不必等整段合成完。

Basic synthesis

Pass text, voice, and output format, and get back audio bytes directly. The default sample rate is 24kHz.

Quality first

tts-1-hd sounds better but has higher latency; suited to pre-generated spoken content, not real-time conversation.

Adjusting speed

speed runs from 0.25 to 4.0, with 1.0 at normal speed. Teaching material is often slowed to around 0.9 to make it easy to follow.

Streaming synthesis

When you need to play while generating, use a streaming format: the client can start playing on the first chunk instead of waiting for the whole clip.