minimax/minimax-voice-clone

Docs
Schema

[Core Function] MiniMax Voice Clone clones a speaker from a public reference audio URL and synthesizes new speech in one async request; only the final audio is returned. [Strengths] No voice enrollment management; optional language_boost for clone and synthesis; prosody controls (speed, volume, pitch, emotion); pronunciation overrides; voice effects; flexible audio formats. [Best For] One-off cloned narration, personalized prompts, and demos where a lasting voice library is not needed. [Limitations] Do NOT use this to obtain a reusable voice library entry; the cloned voice is temporary and is not returned. audio_url must be publicly reachable (mp3/m4a/wav, 10s–5min, ≤20 MB). [Routing] Choose speech-2.8-hd for higher quality; speech-2.8-turbo for lower latency or cost. Paralinguistic tags such as (laughs) in text are supported on both speech-2.8-hd and speech-2.8-turbo.

$69.0000~$115.0000/M chars
speech-to-speech

Input

Text to synthesize with the temporarily cloned voice. Required. Fewer than 10,000 Unicode characters. On speech-2.8-hd / speech-2.8-turbo, may include paralinguistic tags such as (laughs), (breath), or (sighs).
Primary reference audio URL for cloning. Must be publicly accessible without auth. Formats: mp3, m4a, wav. Duration 10 seconds–5 minutes. File size ≤20 MB. Distinct from prompt_audio.
Hint: Drag and drop files, paste from clipboard (Ctrl/Cmd+V), or provide a URL.
Supported MiniMax speech model for both voice clone and TTS in this request.
model
Optional transcript matching prompt_audio. Must align with the example audio content; end with punctuation. Not a substitute for text.
Custom pronunciation rules. Use / as separator. For Mandarin entries, tones are digits 1-5 (5=neutral), e.g. name/(yan4)(shao3)(fei1). Example: ["omg/oh my god"].
Whether to append an AIGC watermark to the returned audio. Default false.
Audio bitrate. Default 128000. Only effective when format is mp3; ignored for other formats.
128000
Output channel count. 1 is mono, 2 is stereo. Default 1.
1
Optional emotion control. Leave unset to let the model infer from text.
emotion
Force constant bitrate encoding. Optional; typically unused for standard async synthesis. Default false.
Output audio encoding format. Default mp3. Result audio is returned as a URL on the completed task.
mp3
Optional language or dialect boost for both cloning and synthesis. Omit for default behavior. Use auto to let the model decide.
language_boost
Enable reading LaTeX formulas aloud (Chinese only). When true, language_boost may be forced to Chinese. Wrap formulas with $$ and escape backslashes. Default false.
Voice effect intensity: lower is stronger, higher is softer. Range [-100, 100].
Voice effect pitch: lower is deeper, higher is brighter. Range [-100, 100]. Distinct from pitch (semitone).
Voice effect timbre: lower is thicker, higher is brighter or crisper. Range [-100, 100].
Whether to enable noise reduction on the clone reference audio. Default false.
Whether to enable volume normalization for cloning. Default false.
Pitch adjustment in semitones. Default 0. Range [-12, 12]. Distinct from modify_pitch (voice effect).
Optional short example audio URL to improve clone similarity and stability. Not a substitute for audio_url. Formats: mp3, m4a, wav. Duration must be under 8 seconds. File size ≤20 MB.
Hint: Drag and drop files, paste from clipboard (Ctrl/Cmd+V), or provide a URL.
Output sample rate in Hz. Default 32000.
32000
Optional sound effect applied to the output. Supported for mp3, wav, and flac.
sound_effects
Speech rate multiplier for synthesis. Default 1.0. Range [0.5, 2.0].
Whether to generate a subtitle download link. Default false.
Enable Chinese/English text normalization for better number reading; may add slight latency. Default false.
Speech volume for synthesis. Default 1.0. Range (0.0, 10.0].

Result

No results yet

Run the model to preview the output here.

Next:

README

Attributes

Collections

Parameters

Name Description Type Required Enums
text Text to synthesize with the temporarily cloned voice. Required. Fewer than 10,000 Unicode characters. On speech-2.8-hd / speech-2.8-turbo, may include paralinguistic tags such as (laughs), (breath), or (sighs). string Yes -
audio_url Primary reference audio URL for cloning. Must be publicly accessible without auth. Formats: mp3, m4a, wav. Duration 10 seconds–5 minutes. File size ≤20 MB. Distinct from prompt_audio. string Yes -
model Supported MiniMax speech model for both voice clone and TTS in this request. string Yes speech-2.8-hd, speech-2.8-turbo
prompt_text Optional transcript matching prompt_audio. Must align with the example audio content; end with punctuation. Not a substitute for text. string No -
pronunciation_tone Custom pronunciation rules. Use / as separator. For Mandarin entries, tones are digits 1-5 (5=neutral), e.g. name/(yan4)(shao3)(fei1). Example: [“omg/oh my god”]. string[] No -
aigc_watermark Whether to append an AIGC watermark to the returned audio. Default false. boolean No true, false
bitrate Audio bitrate. Default 128000. Only effective when format is mp3; ignored for other formats. integer No 32000, 64000, 128000, 256000
channel Output channel count. 1 is mono, 2 is stereo. Default 1. integer No 1, 2
emotion Optional emotion control. Leave unset to let the model infer from text. string No happy, sad, angry, fearful, disgusted, surprised, calm
force_cbr Force constant bitrate encoding. Optional; typically unused for standard async synthesis. Default false. boolean No true, false
format Output audio encoding format. Default mp3. Result audio is returned as a URL on the completed task. string No mp3, pcm, flac, wav
language_boost Optional language or dialect boost for both cloning and synthesis. Omit for default behavior. Use auto to let the model decide. string No Chinese, Chinese,Yue, English, Arabic, Russian, Spanish, French, Portuguese, German, Turkish, Dutch, Ukrainian, Vietnamese, Indonesian, Japanese, Italian, Korean, Thai, Polish, Romanian, Greek, Czech, Finnish, Hindi, Bulgarian, Danish, Hebrew, Malay, Persian, Slovak, Swedish, Croatian, Filipino, Hungarian, Norwegian, Slovenian, Catalan, Nynorsk, Tamil, Afrikaans, auto
latex_read Enable reading LaTeX formulas aloud (Chinese only). When true, language_boost may be forced to Chinese. Wrap formulas with $$ and escape backslashes. Default false. boolean No true, false
modify_intensity Voice effect intensity: lower is stronger, higher is softer. Range [-100, 100]. integer No -
modify_pitch Voice effect pitch: lower is deeper, higher is brighter. Range [-100, 100]. Distinct from pitch (semitone). integer No -
modify_timbre Voice effect timbre: lower is thicker, higher is brighter or crisper. Range [-100, 100]. integer No -
need_noise_reduction Whether to enable noise reduction on the clone reference audio. Default false. boolean No true, false
need_volume_normalization Whether to enable volume normalization for cloning. Default false. boolean No true, false
pitch Pitch adjustment in semitones. Default 0. Range [-12, 12]. Distinct from modify_pitch (voice effect). integer No -
prompt_audio Optional short example audio URL to improve clone similarity and stability. Not a substitute for audio_url. Formats: mp3, m4a, wav. Duration must be under 8 seconds. File size ≤20 MB. string No -
sample_rate Output sample rate in Hz. Default 32000. integer No 8000, 16000, 22050, 24000, 32000, 44100
sound_effects Optional sound effect applied to the output. Supported for mp3, wav, and flac. string No spacious_echo, auditorium_echo, lofi_telephone, robotic
speed Speech rate multiplier for synthesis. Default 1.0. Range [0.5, 2.0]. number No -
subtitle_enable Whether to generate a subtitle download link. Default false. boolean No true, false
text_normalization Enable Chinese/English text normalization for better number reading; may add slight latency. Default false. boolean No true, false
vol Speech volume for synthesis. Default 1.0. Range (0.0, 10.0]. number No -

Pricing

Unit: $/M chars

Dimension Pricing
model: speech-2.8-hd 115.0000
model: speech-02-hd 115.0000
model: speech-2.8-turbo 69.0000
model: speech-02-turbo 69.0000
  • alibaba/cosyvoice-clone: [Core Function] CosyVoice Clone clones a speaker from a public reference audio URL and synthesizes new speech in one async request; only the final audio is returned. [Strengths] No voice enrollment management; language_hint applies to both cloning and synthesis; SSML, hot_fix, and prosody controls. [Best For] One-off cloned narration and demos where a lasting voice library is not needed. [Limitations] Do NOT use this to obtain a reusable voice library entry; the cloned voice is temporary and is not returned. Reference URL must be publicly accessible. model must be cosyvoice-v3.5-plus or cosyvoice-v3.5-flash. [Routing] Choose cosyvoice-v3.5-plus for higher speech quality; cosyvoice-v3.5-flash for lower latency.

Related Resources

Voice Cloning
Voice CloningCollection
alibaba/cosyvoice-clone
alibaba/cosyvoice-clonespeech-to-speech