xenial-black•3d ago

Inefficient algorithm for MiniMax TTS

Hi, The way VAPI handles TTS from minimax will hit the rate limit on almost all their plans within a very short time. I have been testing MiniMax with VAPI, and I can’t even get a call to last a few seconds without hitting the rate limit. The issue seems to be that VAPI is constantly hitting minimax for fragments of a sentence, and instead of just sending the entire sentence and streaming the result, it splits up the sentence into multiple parts that will then be streamed? Can this be changed in anyway? As it stands, it does 10+ requests in <4 seconds of call time, which is a little aggressive imo.

4 Replies

Vapi Support Bot•3d ago

To help reduce MiniMax TTS rate limiting, enable voice response caching in your configuration. This will cache repeated phrases and minimize duplicate requests:

{
  "voice": {
    "provider": "minimax",
    "cachingEnabled": true
  }
}

{
  "voice": {
    "provider": "minimax",
    "cachingEnabled": true
  }
}

Set cachingEnabled to true in your MiniMax voice config to reduce API calls for repeated text (documentation). Source: - Voice Enhancements & Minimax Improvements

⤵️ Mention @Vapi Support Bot for follow-ups 🐥

xenial-blackOP•21h ago

This doesn’t solve the issue right now VAPI seems to be chunking the assistants speech on every period, which is not what I am expecting. Even when I adjust the punctuation boundaries to not include periods, it still will split it up by every period in the sentence adjusting the cachingEnabled and chunkPlan do nothing as well @Vapi Team can I pleae get some input on this? Any updates on this?

Vapi•12h ago

you can also try disabling chunking completely:

{
  "voice": {
    "provider": "minimax",
    "cachingEnabled": true,
    "chunkPlan": {
      "enabled": false
    }
  }
}

{
  "voice": {
    "provider": "minimax",
    "cachingEnabled": true,
    "chunkPlan": {
      "enabled": false
    }
  }
}

xenial-blackOP•8h ago

disabling or enabling chunking doesnt seem to do anything

Inefficient algorithm for MiniMax TTS

Did you find this page helpful?