Written for Yogstation13

Written in nestjs

Uses piper, which can be found in ./piper-src

Piper voice models are individually licensed, please see MODEL_CARD under each voice folder (located in ./piper-voices) to view attributions and licensing.

For additional voices see here.

For a model training guide & video tutorial see here.

You will require Docker

To compile & start the server, run docker compose up --build

Exposes on port 8133

GET

/ping

Requires no auth

returns "Hello, world!"

POST

/tts_clear_cache

Auth key should be send in header Authorization : mysecurekey

Clears out the backend cache of voices, should be used at the start of each round. Cache is expected to reach up to 1GB on particularly active rounds.

GET

/tts?model={desired_model}&pitch={multiplier}

Auth key should be send in header Authorization : mysecurekey

Message and filters should be in JSON body {message: "Hello, world!", "filters":{"robotic":["roundstart"]}

Models can be found in ./piper-voices and should be queried as country-name i.e. GB-alba

Example usage: /tts?model=US-joe&pitch=1