primer · voice

Four Fish Audio primitives, demystified.

Every idea in this repo leans on one of four Fish Audio primitives. Each one runs against the public Fish Audio API with a single key, so you ship a real voice demo with zero infra and nothing to host.

The one-key protocol

Add this in your Lovable project (Settings → Secrets) before pasting any mega-prompt:

FISH_AUDIO_API_KEYSingle API key that unlocks TTS, voice agents, transcription, and voice design. Free tier covers a hackathon weekend.open ↗

streaming voice

Fish Audio Text-to-Speech

250 ideas use this

kernel

a POST to https://api.fish.audio/v1/tts with the `model: s2.1-pro` header that returns MP3 / WAV / PCM / Opus bytes and plays them as the user reads or interacts

how it drives the UI

a 'play' button (or auto-play) that streams lifelike narration of any text the app generates

EXAMPLE IDEAS

live conversation

Voice Loop Agent

250 ideas use this

kernel

a full spoken round trip — browser MediaRecorder captures the question, /v1/asr transcribes it, Lovable AI answers it, /v1/tts speaks the answer back

how it drives the UI

a microphone button that opens a back-and-forth spoken exchange with an assistant tuned for the discipline

EXAMPLE IDEAS

live transcription

Speech-to-Text with Timestamps

250 ideas use this

kernel

Fish Audio /v1/asr with `ignore_timestamps=false` — the client posts recorded audio chunks as multipart form data and gets back text plus per-segment start/end times

how it drives the UI

a caption strip that fills in as the user speaks, with timestamped lines saved on every pause

EXAMPLE IDEAS

custom voices

Voice Design & Cloning

250 ideas use this

kernel

Fish Audio /v1/voice-design (model header `voice-design-1`) invents candidate voices from a written description, and POST /model clones a voice from a reference clip for reuse via `reference_id`

how it drives the UI

a description box that auditions several generated voices, then speaks the app's content in the one the user picks

EXAMPLE IDEAS