IndexTTS Added to External TTS Engines โ
Set engine to index-tts inside ExtParam and IndexTTS runs on the existing TTS APIs. With this addition, MPS now offers six external TTS engines.
- Select the engine: pass
{"engine": "index-tts"}inExtParam(a JSON string). UsesynExt.modelto pickindex-tts-2(the default) orindex-tts-1.5. - Select the voice: put the reference audio URL straight into
VoiceId. There are no voice names and no registration step โ zero-shot cloning is the default behaviour. - Select the emotion: under
engineCustom, choose one of an 8-dimension vector (emo_vector), reference audio (emo_audio_url+emo_alpha), a text description (emo_text), or random sampling (use_random).
It is the only engine that can take voice and emotion from two different clips:
VoiceIddecides who is speaking,emo_audio_urldecides with what emotion.The emotion parameters are supported only by
index-tts-2; selectingindex-tts-1.5makes them silently no-ops.It is slow, so the async API is recommended. Three APIs are supported:
TextToSpeech,TextToSpeechAsyncandDescribeTextToSpeechAsyncTask.Audio parameters such as
format,sampleRateandspeedare not declared in the integration guide. Do not copy another engine's parameter table โ test with short text first. On the output side,ExtInfo.durationreturns the total length in seconds.