Fish Audio
Expressive Text-to-Speech and Voice Cloning
4.7•15 reviews•2.1K followers
Expressive Text-to-Speech and Voice Cloning
4.7•15 reviews•2.1K followers
Fish Audio is the most expressive and emotionally rich text-to-speech model. It generates lifelike voices that capture emotion, rhythm, and nuance with remarkable realism. Fish Audio Voice Clone recreates a natural voice from just 10 seconds of audio—preserving accent, tone, and speaking habits. Proudly built by the open-source team behind So-VITS-SVC and Bert-VITS2, giving a soul to every voice.
This is the 6th launch from Fish Audio. View more
Fish Audio Realtime STT
Launching today
Fish Audio Realtime Speech to Text (transcribe-stream-1, in preview) streams audio over a WebSocket and returns transcripts turn by turn. It detects when the user has finished, tells speech from background noise, and recognizes backchannels like “mm-hm”. Per the docs, first draft text arrives about 0.5 seconds after speech starts. Supports 29 language codes, custom vocabulary, speaker tags and 8 to 48 kHz PCM audio. $0.36 per hour.
Launch tags:Artificial Intelligence
Launch Team
Framer AI AgentsDesign and publish professional sites with AI
Promoted
Fish Audio Realtime Speech to Text is a live transcription API for voice agents, running on transcribe-stream-1. The docs say it’s in preview.
You stream audio over a WebSocket and get transcripts back turn by turn. Besides the text, it works out when the user has actually finished talking, which is the hard part of building a voice agent that doesn’t cut people off or wait too long.
Here’s what that covers:
live draft text while the person is still speaking, then one final transcript per turn
an “eager end of turn” signal, so your agent can start its LLM call early
backchannel detection, so an “mm-hm” while the agent talks isn’t treated as an interruption
background awareness, so a TV or side conversation usually comes back marked as not the user’s turn
custom vocabulary (up to 100 terms), prior conversation context, and speaker tags
29 language codes, and 16-bit PCM audio from 8 to 48 kHz (phone audio needs decoding to PCM first)
On speed, the docs say the first draft arrives about 0.5 seconds after speech starts, and the final transcript about 1.1 seconds after the speaker stops. Those were measured on production with synthetic speech, and network and audio sending add to them.
Pricing is $0.36 per hour, billed per second of session time. Sessions can run up to 60 minutes.
Good fit if you’re building a voice agent, live captions or phone-call transcription and want turn detection included instead of building it yourself. It’s developer-only, with no SDK for this endpoint, but the cookbook has tested Python and Node.js clients.
Realtime Speech to Text docs