MiniMax

MiniMax generates speech with its Speech models in more than twenty languages, including English, Mandarin, Cantonese, Japanese, Korean and most European languages. MiniMax tuning is passed in synthesizer.options.

{
  "verb": "say",
  "text": "Hi there, your order ships today.",
  "synthesizer": {
    "vendor": "minimax",
    "language": "en-US",
    "voice": "English_expressive_narrator",
    "options": {
      "speed": 1.1,
      "emotion": "happy"
    }
  }
}

You can also set these in the portal under Extra Options on the speech credential, in which case they apply to every request that uses that credential. Settings supplied on the verb take precedence.

Credential

Create an API key in the MiniMax developer platform. jambonz sends the key as a Bearer token; no Group ID is needed. Add it as a MiniMax speech credential in the portal or API. MiniMax is used for text-to-speech only.

Models

Model	Notes
`speech-2.8-turbo`	Default. Low latency.
`speech-2.8-hd`	Higher quality, more latency.
`speech-2.6-turbo`, `speech-2.6-hd`	Previous generation.
`speech-02-turbo`, `speech-02-hd`	Older generation.
`speech-01-turbo`, `speech-01-hd`	Oldest generation.

Pick the model on the speech credential, or set options.model_id to override it for one verb.

Voices and languages

A voice is a MiniMax voice id such as English_expressive_narrator or Japanese_IntellectualSenior, sent exactly as written. The portal lists MiniMax’s system voices by language, taken from MiniMax’s voice list. Cloned or designed voices in your MiniMax account can be used by their voice id.

The voice alone selects the language: synthesizer.language is not sent to MiniMax. To help MiniMax with a language or dialect, set options.language_boost, for example English, Chinese,Yue or auto.

Options

model_id

string

The Speech model to use, overriding the credential’s model.

speed

number

Speaking rate, 0.5 – 2. MiniMax’s default is 1.

vol

number

Volume, above 0 up to 10. MiniMax’s default is 1.

pitch

integer

Pitch shift in semitones, -12 – 12. MiniMax’s default is 0.

emotion

string

happy, sad, angry, fearful, disgusted, surprised, calm, fluent or whisper. By default MiniMax picks the emotion from the text.

text_normalization

boolean

Read numbers, dates and symbols as words in Chinese and English.

english_normalization

boolean

Improve how numbers are read in English. Streaming only.

latex_read

boolean

Read LaTeX formulas written between $ signs.

language_boost

string

The language or dialect to favour, by MiniMax’s name for it, such as English, Chinese,Yue or auto.

text_normalization_mode

string

basic or quality.

continuous_sound

boolean

Smooth the joins between sentences. speech-2.8-* models and streaming only.

voice_modify

object

Voice effects: pitch, intensity and timbre, each -100 – 100, and sound_effects, one of spacious_echo, auditorium_echo, lofi_telephone or robotic. MiniMax applies these to non-streamed audio only, so they have no effect on streamed speech.

pronunciation_dict

object

Custom pronunciations, as {"tone": ["omg/oh my god"]}.

timbre_weights

array

Mix up to four voices, as [{"voice_id": "English_expressive_narrator", "weight": 1}].

Every option is optional. An option you don’t set is not sent, so MiniMax’s own default applies.

Endpoint override

To point at a non-default endpoint, such as MiniMax’s api-uw.minimax.io or api.minimaxi.com, or a proxy, set api_uri on the credential or in its Extra Options. Leave it out to use api.minimax.io.

Streaming

MiniMax supports TTS streaming, so it can be driven token by token from an LLM using tts:tokens, for example from the agent verb. jambonz streams text to MiniMax over a live connection, and on barge-in sends MiniMax’s cancel instead of reconnecting. jambonz does not request MiniMax’s word timestamps, so alignment events are not available.