Fish Audio

Fish Audio (API docs) is a voice provider for the voice clone, picked with --provider fishAudio or LOG_VOICE_PROVIDER=fishAudio (rule 143). It was the first provider (rule 142). Creating a voice model through its API needs API credit: on 2026-11-05 it answered 403, "API model creation limit reached (0/0)", so Inworld became the default.

The three calls

The client (src/extensions/fishAudio/client.ts) knows only these, each with FISH_API_KEY as a bearer token:

• POST https://api.fish.audio/model, multipart: type tts, title, train_mode fast, visibility private, and the voices files (matt's own voice notes as they are, 10 to 60 s). The answer's _id is the voice id.
• GET https://api.fish.audio/model/{id}: the model's state.
• POST https://api.fish.audio/v1/tts, JSON text, reference_id, format, with the model header: s2.1-pro-free ($0, the default) or s2.1-pro ($15 per million UTF-8 bytes, only when --model names it).

Setting it up

• FISH_API_KEY in .env.local; its name is in .env.example.
• log clone voice --provider fishAudio --apply once, when the account has credit (matt's call, rule 41).