fish-audio-sdk

Fish Audio SDK Skill

Use this skill to generate correct, runnable code with the official Fish Audio SDKs:

• Python: package fish-audio-sdk on PyPI, imported as fishaudio. (The same wheel still ships a separate legacy fish_audio_sdk package. Do not mix them; everything here is the modern fishaudio package.)
• JavaScript / TypeScript: package fish-audio on npm, imported as FishAudioClient.

If the user wants raw curl / HTTP / WebSocket without installing an SDK, use the fish-audio-api skill instead.

This file is the index. Deeper, task-specific rules and full examples live in references/. Read the reference for the task you're doing before writing code.

Global facts

• Auth: both SDKs read the API key from the FISH_API_KEY environment variable automatically. Get keys at https://fish.audio/app/api-keys. Never hardcode a key.
• Base URL: https://api.fish.audio (override with base_url= in Python / baseUrl: in JS).
• TTS models: the API supports s1, s2-pro, s2.1-pro (recommended for production), and s2.1-pro-free (free tier), but the SDK type definitions currently list only s1 and s2-pro (s2-pro = SDK default). Both SDKs forward the model value without runtime validation, so "s2.1-pro" works over the wire. Static type checkers will flag it, so add # type: ignore (Python) / an as cast (TS), or use the fish-audio-api skill for raw calls. speech-1.5 / speech-1.6 are deprecated. In Python pass model="s2-pro" (keyword); in JS pass the positional backend argument.
• ASR models: use transcribe-1-pro (recommended: speaker turns, long recordings, emotion cues). Neither SDK has an ASR model argument: send the model: transcribe-1-pro HTTP header on every request (Python RequestOptions(additional_headers=...), JS requestOptions.headers). A request without it is served and billed as transcribe-1, the model for short recordings. See references/speech-to-text.md.
• Audio formats: mp3 (default), wav, pcm, opus.
• Playback in examples: play() shells out to a system audio tool: Python uses ffmpeg/ffplay (or mpv), JS uses ffplay. It is for local/desktop use; in a server, save() to a file or stream the bytes instead. See references/installation.md.

Quick start: Python

from fishaudio import FishAudio
from fishaudio.utils import play, save

client = FishAudio()  # reads FISH_API_KEY

# Generate speech (returns the full audio as bytes)
audio = client.tts.convert(text="Hello from Fish Audio!")

save(audio, "output.mp3")   # write to a file
# play(audio)               # or play locally (needs ffmpeg)

Async: identical resource tree on AsyncFishAudio, used as a context manager:

import asyncio
from fishaudio import AsyncFishAudio
from fishaudio.utils import save

async def main():
    async with AsyncFishAudio() as client:
        audio = await client.tts.convert(text="Hello from Fish Audio!")
        save(audio, "output.mp3")

asyncio.run(main())

Quick start: JavaScript / TypeScript

import { FishAudioClient, play } from "fish-audio";

const client = new FishAudioClient({ apiKey: process.env.FISH_API_KEY });

// convert() returns audio you can play or pipe to a file
const audio = await client.textToSpeech.convert({
  text: "Hello from Fish Audio!",
}); // defaults to model "s2-pro"
await play(audio); // local playback (needs ffplay)

To pick a model in JS, pass backend as the positional argument (not a named option):

const audio = await client.textToSpeech.convert({ text: "Hi" }, "s1");

Capabilities → references

Task	Reference
Install, auth, playback deps, verify a key	references/installation.md
Text-to-Speech (convert, stream, formats, prosody, model select)	references/text-to-speech.md
Voice cloning (instant references + persistent voice models)	references/voice-cloning.md
Speech-to-Text (models, timestamps, fields the SDK drops)	references/speech-to-text.md
Realtime WebSocket TTS (stream text → audio)	references/websocket.md
Errors, retries, and timeouts (the real exception types)	references/errors.md

Python ↔ JavaScript name map

The two SDKs do not use the same names. Use this map when porting code between them.

Concept	Python (fishaudio)	JavaScript (fish-audio)
Client	FishAudio() / AsyncFishAudio()	new FishAudioClient({ apiKey })
Text-to-Speech	client.tts.convert(text=...) → bytes	client.textToSpeech.convert({ text })
TTS HTTP stream	client.tts.stream(...) → AudioStream	(use convert; realtime streaming is convertRealtime)
Realtime WebSocket	client.tts.stream_websocket(text_stream)	client.textToSpeech.convertRealtime(request, textStream)
Speech-to-Text	client.asr.transcribe(audio=...)	client.speechToText.convert({ audio })
List voice models	client.voices.list()	client.voices.search()
Get voice model	client.voices.get(id)	client.voices.get(id)
Create voice (clone)	client.voices.create(title=..., voices=[bytes])	client.voices.ivc.create({ title, voices: [File] })
Update / delete voice	client.voices.update(id, ...) / delete(id)	client.voices.update(id, ...) / delete(id)
Credit balance	client.account.get_credits()	client.user.get_api_credit()
Subscription package	client.account.get_package()	client.user.get_package()
Choose model	model="s2-pro" keyword arg	positional backend arg, e.g. convert(req, "s2-pro")
ASR model (header)	RequestOptions(additional_headers=...)	convert(req, { headers: { model: "transcribe-1-pro" } })

Decision shortcuts

• Audio from text → tts.convert (Python) / textToSpeech.convert (JS).
• Reuse a saved voice → pass reference_id (the voice model id).
• Clone a voice instantly from a clip → pass references=[ReferenceAudio(audio=..., text=...)] (Python) / references: [{ audio, text }] (JS). See voice-cloning.
• Persistent custom voice to reuse → create a voice model, then use its id as reference_id.
• Stream tokens from an LLM and play speech as it arrives → tts.stream_websocket (Python) / textToSpeech.convertRealtime (JS). See websocket.
• Transcribe audio → asr.transcribe (Python) / speechToText.convert (JS) with the model: transcribe-1-pro header (recommended; without it, the request runs on transcribe-1). Pro request fields, speaker_turns, request_id, and the language fields need raw HTTP in Python (fish-audio-sdk 1.3.0). See speech-to-text.

Gotchas (verified against the SDK source)

• Python latency accepts only "normal" or "balanced" (default "balanced"); there is no "low".
• The Python client has no max_retries and does not auto-retry; the JS client does auto-retry (configurable via per-call requestOptions.maxRetries). See errors.
• Python defines a ValidationError class but never raises it, so don't catch it expecting validation failures; a 422 surfaces as APIError. The JS SDK throws UnprocessableEntityError on 422.
• ASR segment start / end and duration are all in seconds. The Python SDK's ASRResponse docstring says milliseconds; that is wrong. See speech-to-text.