Skip to content

TTS-MiniMax ​

Updated

This page covers calling MiniMax speech synthesis through MPS. MiniMax is integrated under the engine code minimax: the APIs, authentication, billing, task polling and storage all reuse the existing MPS speech synthesis APIs β€” you only set "engine": "minimax" inside the ExtParam field to route to MiniMax, then add the model, emotion, speed and other parameters as needed.

MiniMax has the broadest API coverage of the external TTS engines: it supports speech synthesis, voice cloning and voice design at the same time, and offers seven emotions (emotion) plus independent volume and pitch control.

Choosing between the engines

  • Need per-character timestamps for karaoke-style subtitles β†’ TTS-ElevenLabs
  • Need to describe accent and tone in one natural-language sentence β†’ TTS-Gemini (engineCustom.prompt)
  • Need scenario broadcast styles (news, advertising, radio, ASMR…) β†’ TTS-Seed
  • Need custom emotion tags, or cloning without creating a voice β†’ TTS-Fishaudio
  • Need enumerated emotions (happy / sad / angry …), independent volume and pitch, or both cloning and design β†’ this page

The full comparison of all seven engines is in MPS TTS Engines Compared.

APIs and official references ​

All APIs below belong to the MPS "Media AI – Dubbing APIs" category. Official API references live under Media AI – Dubbing APIs on Tencent Cloud International.

APIs that support minimax ​

The MiniMax engine integration guide explicitly lists these 7 APIs as switchable to MiniMax via ExtParam.engine:

APIPurposeCall typeOfficial reference
TextToSpeechShort text to speech, returns audio directlySync1041/82311
CloneVoiceClone a voice from reference audioSync1041/82312
TextToSpeechAsyncLong text to speech, returns a TaskIdAsync1041/79693
DescribeTextToSpeechAsyncTaskPoll async synthesis result (Status / AudioUrl)Sync1041/79696
DesignVoiceAsyncDesign a new voice from a text description, returns a TaskIdAsync1041/79694
DescribeDesignTaskPoll voice design task resultSync1041/79598
DescribeVoicesQuery available voices (including MiniMax system voices)Sync1041/79695

Other APIs in the same category ​

APIPurposeWorks with MiniMaxOfficial reference
UpdateVoice / DeleteVoiceUpdate / delete a cloned or designed voice⏳ Not listed in the guide1041/81214 / 1041/81215
ChangeVoiceVoice conversion⏳ Not listed in the guide1041/83750
SyncDubbingLegacy combined API (clone or synthesize)⏳ Not listed in the guide1041/77775

⏳ = not mentioned in the integration guide; whether it can be switched to MiniMax should be confirmed by actual calls.

Shared conventions:

  • Version 2019-06-12; endpoint mps.intl.tencentcloudapi.com on International (mps.tencentcloudapi.com in the Chinese mainland).
  • Authentication is TC3-HMAC-SHA256 (SecretId / SecretKey); see Quick Start for the signing method.
  • Rate limit is 20 requests/second per API.

Comparison with the other engines ​

MPS currently offers seven external TTS engines. The full side-by-side comparison (API coverage, emotion mechanism, audio parameters, cloning, selection guidance) is on the Blog: MPS TTS Engines Compared β€” the comparison table is maintained only there, so the engine pages cannot drift apart.

Where minimax sits: the broadest API coverage β€” synthesis, cloning and voice design are all supported, with independent control over emotion, volume and pitch.

Engine switching: ExtParam.engine ​

The MPS speech synthesis APIs are engine-agnostic by themselves β€” the ExtParam JSON string decides which engine runs the request. Parameters come in three layers:

LayerPurposeNotes
engineSelects the engineAlways minimax on this page
synExt.*MPS-normalized synthesis parametersModel, emotion, format, speed β€” field names are unified across engines
synExt.engineCustom.*Pass-through to MiniMax's native parametersSupports every parameter of the MiniMax official request body

The outer layer wins

When the same key appears in both synExt.xxx and synExt.engineCustom.xxx, the outer synExt value takes effect. To use a MiniMax-only parameter, put it in engineCustom only β€” do not repeat it in synExt.

ExtParam is a string in the request, not an object β€” the JSON must be escaped and passed as a string (see Examples).

Key minimax parameters ​

ParameterTypeRequiredDescription
engineStringYesSet to minimax
synExtObjectNoSynthesis extension parameters
synExt.modelStringNoModel; supports speech-2.6-hd (default), speech-2.8-hd and others
synExt.emotionStringNoEmotion; see Emotion values
synExt.formatStringNoAudio format: mp3, pcm, flac, wav, pcmu_raw, pcmu_wav, opus; default wav
synExt.sampleRateIntegerNoSample rate: 8000, 16000, 22050, 24000, 32000, 44100; default 16000
synExt.durationFloatNoTarget audio duration
synExt.speedFloatNoSpeech rate, range [0.5, 2], default 1.0
synExt.volFloatNoVolume, range (0, 10], default 1.0
synExt.pitchIntegerNoPitch, range [-12, 12], default 0 (original pitch)
synExt.cutSilenceBoolNoTrim leading/trailing silence; currently Chinese and English only
synExt.engineCustomObjectNoEngine-specific parameters. Passes through all official request parameters of the engine; the outer layer wins on conflicts

Emotion values ​

synExt.emotion has seven values and is MiniMax's main differentiator from the other two engines:

ValueMeaning
happyHappy
sadSad
angryAngry
fearfulFearful
disgustedDisgusted
surprisedSurprised
calmNeutral

How emotion, pitch and volume divide the work

emotion sets the emotional baseline for the whole clip, while pitch and vol are physical adjustments layered on top of it. For a "low and calm" narration, emotion: calm plus pitch: -3 sounds more natural than tuning pitch alone.

Voices (VoiceId) ​

For MiniMax, VoiceId takes a MiniMax voice name directly, for example Chinese (Mandarin)_News_Anchor. Available voices can be queried with DescribeVoices β€” see Querying system voices.

Parameter example ​

A complete ExtParam object (expanded for readability; it must be escaped into a string when sent):

json
{
  "engine": "minimax",
  "synExt": {
    "model": "speech-2.6-hd",
    "emotion": "happy",
    "speed": 0.9,
    "engineCustom": {
      "pronunciation_dict": {
        "tone": [
          "倄理/(chu3)(li3)",
          "東京/γƒˆγ‚¦γ‚­γƒ§γ‚¦"
        ]
      },
      "subtitle_enable": false
    }
  }
}

The two engineCustom fields here are MiniMax-only capabilities:

  • pronunciation_dict.tone: custom pronunciation. 倄理/(chu3)(li3) uses toned pinyin to pin down a heteronym, while 東京/γƒˆγ‚¦γ‚­γƒ§γ‚¦ uses kana to specify the reading of Japanese kanji.
  • subtitle_enable: whether to output a subtitle file.

Examples ​

Speech synthesis β€” TextToSpeech ​

json
{
  "Text": "您ε₯½",
  "VoiceId": "Chinese (Mandarin)_News_Anchor",
  "ExtParam": "{\"engine\":\"minimax\",\"synExt\":{\"model\":\"speech-2.6-hd\",\"emotion\":\"happy\",\"speed\":0.9}}"
}

Note that the value of ExtParam is an escaped JSON string, not a nested object.

Voice cloning β€” CloneVoice ​

json
{
  "AudioData": "base64-encoded audio",
  "ExtParam": "{\"engine\":\"minimax\"}"
}

Cloning only needs the engine; synExt is unnecessary β€” synthesis parameters are supplied later when calling TextToSpeech.

Async synthesis and polling ​

For long text, submit with TextToSpeechAsync β€” the ExtParam format is identical to the sync API β€” and it returns a TaskId. Then poll Status with DescribeTextToSpeechAsyncTask and read the audio from AudioUrl once it completes.

Querying system voices ​

Use DescribeVoices (1041/79695) to query the voices available for MiniMax. ExtParam only needs the engine:

json
{
  "engine": "minimax"
}

Here too, ExtParam is passed as a JSON string. The API also supports filtering; common parameters:

ParameterTypeDescription
VoiceTypeStringsystem / clone / design / all (default)
GenderStringmale / female / unknown
AgeStringchild / teenager / youth / middle_aged / senior / unknown
Languages.NArray of StringLanguage
Labels.NArray of StringTag
PageNum / PageSizeIntegerPagination; leaving PageSize empty returns all matching data

Finding your own cloned voices

Cloned and designed voices are queried through the same API β€” set VoiceType to clone or design. Omitting it defaults to all, which returns everything.

Limits and gotchas ​

  • ExtParam is a string: it must be an escaped JSON string in the request; passing an object is rejected as a parameter error.
  • The outer layer wins: synExt.xxx overrides synExt.engineCustom.xxx β€” never write the same parameter in both layers.
  • cutSilence is Chinese/English only: it has no effect for other languages.
  • The default format is wav, not mp3 β€” set format explicitly when size matters.
  • The default sample rate is 16000: request 44100 explicitly for music or high-fidelity scenarios.
  • Volume is an open interval (0, 10]: 0 is invalid; the minimum must be greater than 0.
  • APIs marked ⏳ should be confirmed by actual calls: the βœ… / ⏳ marks follow what the MiniMax engine integration guide lists, and an unlisted API is not necessarily unavailable.
DocumentNotes
MPS TTS Engines ComparedCapabilities, parameters and selection guidance across all seven external engines
AI DubbingOverview of MPS dubbing and its built-in engines
TTS-ElevenLabsThe ElevenLabs engine (tts-l1) on the same APIs, with per-character timestamps
TTS-GeminiThe Google Gemini-TTS engine (tts-gg) on the same APIs, style controlled by prompt, no cloning
TTS-SeedThe TTS-Seed engine (tts-s) on the same APIs, with 24 emotion values
TTS-FishaudioThe Fish Audio engine (tts-f) on the same APIs, with inline emotion tags and zero-shot cloning
TTS-IndexThe same APIs on IndexTTS (index-tts): a reference audio URL straight in VoiceId, 8-dimension emotion vector
TTS-SeedaudioThe same APIs on Seed Audio (tts-sa): the only engine that generates speech from an image, word-level timestamps
MiniMax MusicMiniMax's music generation capability β€” a different API from the speech synthesis here
Quick StartAuthentication and request signing
Media AI – Dubbing APIsOfficial API category

Built with VitePress Β· Deployed on Tencent EdgeOne Pages