# AnythingLLM text to speech and transcription

An AnythingLLM workspace used in more than one language needs replies read aloud in each of them. Choose OpenAI Compatible as AnythingLLM's text-to-speech provider and set Paxa's base URL and an API key. Replies then play in English, Thai, and Mandarin, and Paxa transcribes dictated prompts and uploaded recordings.

## Each provider, pointed at Paxa

Speech, voice input, and uploads each have their own provider, and each takes the same base URL and key. A Docker or bare-metal deployment sets them all in .env.

## From a key to a workspace that speaks

AnythingLLM keeps its LLM, documents, and agents. Paxa reads replies aloud and transcribes the audio you give it.

1. **Read replies aloud**: In Settings > Voice & Speech, choose OpenAI Compatible under Text-to-speech Preference. Set Base URL to https://api.paxalabs.com/v1, API Key to your Paxa key, TTS Model to paxa-tts-flash-v2, and Voice Model to a Paxa voice id.
2. **Dictate prompts**: On the same page, choose Generic OpenAI under Speech-to-text Preference, with the same base URL and key and Transcription Model set to paxa-stt-lite-v1-preview.
3. **Transcribe uploads**: In Settings > Transcription, choose OpenAI Compatible and set the same Base URL, API Key, and Transcription Model. Left empty, the model falls back to one Paxa refuses.

Paxa TTS Flash v2 lists at $10 per 1M characters. Paxa STT Lite lists at $0.50 per hour of audio, billed by the millisecond.

## Questions

### What is AnythingLLM?

AnythingLLM describes itself as an all-in-one AI application that can do RAG, AI Agents, and much more. It runs as a desktop app and in Docker, and each of its voice features takes an OpenAI-compatible provider.

### Which voices can AnythingLLM use?

AnythingLLM can use any of the 50 Paxa TTS Flash v2 voices, and each one speaks English, Thai, and Mandarin. Left empty, the model and voice fall back to `tts-1` and `alloy`, which Paxa also accepts. [The voice library](https://paxalabs.com/voices) plays every voice.

### Why does an uploaded recording fail with unknown_model?

Left empty, the Transcription Model under Settings > Transcription falls back to a Whisper model name Paxa does not serve, and Paxa answers [400 unknown_model](https://paxalabs.com/docs/errors#unknown_model). Set it to `paxa-stt-lite-v1-preview`.

### How long can an uploaded recording run?

A recording can run up to 120 minutes and 95 MiB. Paxa STT Lite is a Research preview.

## Start with free credits

Sign in with Google, GitHub, or Hugging Face and spend 100 one-time free credits on your own text.