APIMaster.ai

MiniMax Speech 2.8 HD / Turbo · API reference

Convert text to audio, select voices and formats, and read buffered SSE responses.

Send a JSON request to POST /v1/wand/minimax-tts/sync_tts with an APIMaster API key in the Authorization: Bearer header. Set model to minimax-speech-2.8-hd or minimax-speech-2.8-turbo.

Field Type Description
text string Required, fewer than 10,000 characters. Use newlines for paragraphs, <#x#> for a pause of x seconds, and parenthesized pinyin or IPA for pronunciation. The models support cues such as (laughs).
voice_setting object Required. Set voice_id to a system voice or to the complete reference returned by voice design or voice cloning. When mixing voices, leave this ID empty and set timbre_weights.
voice_setting.speed, voice_setting.vol, voice_setting.pitch number Speed 0.5–2 (default 1). Volume greater than 0 and at most 10 (default 1). Integer pitch −12 to 12 (default 0).
voice_setting.emotion string happy, sad, angry, fearful, disgusted, surprised, calm, or fluent. Omit to let the model choose.
voice_setting.text_normalization, voice_setting.latex_read boolean Both default to false. Wrap equations in $$ for LaTeX reading.
audio_setting.sample_rate integer 8000, 16000, 22050, 24000, 32000 (default), or 44100 Hz.
audio_setting.bitrate integer MP3 only. 32000, 64000, 128000 (default), or 256000 bps.
audio_setting.format string mp3 (default), pcm, flac, wav, pcmu_raw, pcmu_wav, or opus.
audio_setting.channel integer 1 (default) or 2.
pronunciation_dict.tone string[] Entries in original/replacement form. Replacements can use text, pinyin, IPA, or Japanese kana.
timbre_weights object[] Up to four entries with voice_id and weight from 1 to 100.
language_boost string Optional language enhancement. Use auto, a supported language, or omit.
voice_modify.pitch, voice_modify.intensity, voice_modify.timbre integer −100 to 100.
voice_modify.sound_effects string spacious_echo, auditorium_echo, lofi_telephone, or robotic. For non-streaming, these effects work with MP3, WAV, and FLAC. For SSE, use MP3.
subtitle_enable, subtitle_type boolean, string Subtitles are off by default. The type is sentence (default) or word.
output_format string hex (default) or url for a non-streaming request. URLs last 24 hours.
aigc_watermark, stream boolean Both default to false. SSE is buffered until generation and billing complete and sends hex audio.

language_boost accepts 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, and Afrikaans.

System voices include English_Graceful_Lady, English_Trustworthy_Man, male-qn-qingse, male-qn-jingying, female-shaonv, and female-yujie. Use a custom voice returned by Voice Design or Voice Clone from the same account.

curl -X POST "https://apimaster.ai/v1/wand/minimax-tts/sync_tts" \
  -H "Authorization: Bearer $APIMASTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"minimax-speech-2.8-hd","text":"Welcome to the program.","voice_setting":{"voice_id":"English_Graceful_Lady"},"output_format":"url"}'

In a non-streaming response, check that base_resp.status_code is 0 and data.status is 2. data.audio contains the hex-encoded audio or its URL, and data.subtitle_file can contain a JSON subtitle file. extra_info can include length in milliseconds, sample rate, size, format, and usage_characters. Save trace_id or request_id if you need help diagnosing a failure.

For SSE, data.status=1 is a segment and 2 is completion, potentially followed by [DONE]. Join hex segments in order and decode them to bytes. Subtitles may contain character offsets, time offsets in milliseconds, and timestamped_words. Use the cumulative usage.total_tokens or usage.total_token once instead of summing cumulative frames.

See the HD and Turbo model cards for current APIMaster prices. Initial synthesis with a custom voice may also include a voice fee.