Text to Speech · IntroductionNEW📋 Copy as Markdown

Text-to-Speech (TTS) API: send text, get a link to the synthesized speech (platform CDN direct link, mono 24kHz), optionally with a precisely aligned SRT subtitle file. Synchronous — one request returns the result directly, no polling and no taskId. 32 curated Chinese / English voices (322 in the full catalog), adjustable rate / pitch / volume, and per-sentence prosody control; long text is auto-segmented and stitched. Three output formats: mp3 / mp3-high / webm. Billed by synthesized audio seconds: charged synchronously, non-refundable, minimum 1 credit per call; subtitles cost nothing extra.

Endpoints

Authentication

Authorization: Bearer <API_KEY>
Content-Type: application/json

Request Parameters

Output formats

Per-sentence control (segments)

Use segments instead of text to give each sentence its own rate / pitch / volume — e.g. a normal-paced opening, a slower and louder key sentence, then a quick wrap-up. Up to 200 segments.

• Fields omitted in a segment inherit the top-level value, so you only write what you want to change.
• All segments share one voice (voice is top-level only). A single request cannot mix voices — use separate requests for multiple characters.
• There is no pause tag between segments — the speech service cannot insert silence. Use punctuation for pauses, or split into multiple requests and stitch them yourself.

curl -X POST https://api.cqtai.com/v1/tts \
  -H 'Authorization: Bearer <API_KEY>' \
  -H 'Content-Type: application/json' \
  -d '{
    "voice": "zh-CN-YunxiNeural",
    "segments": [
      { "text": "先说一句正常语速的开场白。" },
      { "text": "这一句要慢下来，重点强调。", "rate": "-25%", "volume": "loud" },
      { "text": "最后快速收尾。", "rate": "+35%", "pitch": "+2st" }
    ]
  }'

Subtitles (SRT)

Add "subtitle": true to the request and the response gains a subtitleUrl. The timeline is exact, not recognized — it comes from the word-boundary positions the speech engine reports while synthesizing, so there are no ASR typos and nothing drifts out of sync.

curl -X POST https://api.cqtai.com/v1/tts \
  -H 'Authorization: Bearer <API_KEY>' \
  -H 'Content-Type: application/json' \
  -d '{
    "text": "第一句话。第二句话稍微长一点，用来演示断句。",
    "voice": "zh-CN-XiaoxiaoNeural",
    "subtitle": true
  }'
# -> { "code":200, "data":{
#      "url":"https://cdn.novapi.ai/tts/....mp3",
#      "subtitleUrl":"https://cdn.novapi.ai/tts/....srt",
#      "duration":6.20, "cost":1, "format":"mp3" } }

• No extra charge and no extra latency (the data falls out of synthesis itself).
• The text matches exactly what you sent (nothing is rewritten by recognition).
• Stays aligned after rate / segments changes the pace (the timeline shifts and the subtitle shifts with it).
• subtitle works with all three formats (the subtitle file is independent of the audio format).
• Use it for a web player’s <track kind="subtitles">, import into an editor, or pass subtitleUrl straight to the video subtitle burn-in endpoint (FFmpeg burn_subtitle).

Layout rules (fixed, not configurable yet): single line; up to ~20 Chinese characters / 40 Latin characters per line; each cue lasts 0.7–7 seconds; breaks at sentence-final punctuation and at pauses longer than 0.6 s. The file is standard SubRip, UTF-8 without BOM, LF line endings.

Request Example

curl -X POST https://api.cqtai.com/v1/tts \
  -H 'Authorization: Bearer <API_KEY>' \
  -H 'Content-Type: application/json' \
  -d '{
    "text": "你好，欢迎使用语音合成服务。",
    "voice": "zh-CN-XiaoxiaoNeural"
  }'
# -> { "code":200, "data":{ "url":"https://cdn.novapi.ai/tts/....mp3", "duration":3.62, "cost":1, "format":"mp3" } }

curl -X POST https://api.cqtai.com/v1/tts \
  -H 'Authorization: Bearer <API_KEY>' \
  -H 'Content-Type: application/json' \
  -d '{
    "text": "This is a faster, higher-pitched sample.",
    "voice": "en-US-AriaNeural",
    "rate": "+15%",
    "pitch": "+3Hz"
  }'

Response Example

{
  "code": 200,
  "msg": "success",
  "data": {
    "url": "https://cdn.novapi.ai/tts/1756272000000_ab12cd.mp3",
    "duration": 3.62,
    "cost": 1,
    "format": "mp3"
  }
}

Response Fields

Supported voices

• The voice list is loaded live from GET /v1/tts/voices (the docs page shows curated voices by default; switch to “Full catalog” for all 322 across 142 locales). voice accepts any shortName listed there.
• For mixed Chinese/English text use AvaMultilingual / AndrewMultilingual — one voice reads both, no per-language switching.
• In ?all=1 the voices array may occasionally be empty (the catalog is still loading right after start-up); loadedAt is then an empty string — fall back to presets rather than treating it as “no voices”.
• The live API response is always authoritative.

curl -X GET https://api.cqtai.com/v1/tts/voices \
  -H 'Authorization: Bearer <API_KEY>'
# -> { "code":200, "data":[ { "id":"edge:zh-CN-XiaoxiaoNeural",
#       "shortName":"zh-CN-XiaoxiaoNeural", "name":"晓晓（温柔女声）",
#       "nameEn":"Xiaoxiao (warm female)", "gender":"female",
#       "language":"zh", "locale":"zh-CN" }, ... ] }

# 全量音色（322 个 / 142 种语言地区）
curl -X GET 'https://api.cqtai.com/v1/tts/voices?all=1' \
  -H 'Authorization: Bearer <API_KEY>'
# -> { "code":200, "data":{ "presets":[...], "voices":[...], "formats":[...], "loadedAt":"..." } }

Capability limits

This service controls voice + rate + pitch + volume. The following are not supported — do not port parameters from other TTS products:

Billing & Credits

• Billed by synthesized audio seconds: credits = max(1, ceil(seconds × unit price)), default 0.1 credit/sec (10s ≈ 1 credit, 60s ≈ 6 credits).
• Charged synchronously, non-refundable: charged once synthesis succeeds and a link is produced; on insufficient balance nothing is charged and no link is returned.
• Subtitles are free: subtitle: true costs exactly the same as without it (the timestamps come out of synthesis itself, nothing is computed twice).
• Parameter errors (400) are returned before synthesis starts and consume no credits.
• Minimum 1 credit per call. Final credits settle at your user rate; a dedicated unit price takes precedence if configured.

Common Errors