Text to Speech APIBuild Voice Into Any App
Expressive, directable speech at ~100ms latency, the same S2.1 Pro model behind HeyGen, Retell, and Plaud.
Running in production at
Models built for different production needs
One API. Switch models with a single header. Both API models bill at the same flat rate.
S2 Pro
Accurate on even the shortest utterances
- Instant voice cloning
- Full multilingual support
- Reliable on very short prompts
- Same flat rate as S2.1 Pro
S2.1 Pro
The most expressive model Fish ships.
- Time-to-first-audio ~100ms (no single H200)
- Real-time factor 0.195
- 80+ languages, instant code-switching
- Open-domain Emotion Tags
- Native multi-speaker in one generation
On-premise
Deploy in your own infrastructure.
- VPC, on-prem, sovereign cloud, air-gapped
- Zero data egress
- Full model control
- Commercial license available
From signup to first audio in 5 minutes.
No sales call required. Get an API key, install the SDK, and ship.
curl https://fish.keatska.top/apifish/v1/tts \-H "Authorization: Bearer $FISH_API_KEY" \-H "Content-Type: application/json" \-H "model: s2.1-pro-free" \-d '{"text": "[chuckle] When you are building something new, there is this [emphasis] mix of wonder and fear.","format": "mp3"}' --output speech.mp3
# pip install fish-audio-sdkfrom fishaudio import FishAudiofrom fishaudio.utils import saveclient = FishAudio() # reads FISH_API_KEYaudio = client.tts.convert(text="Hello from Fish Audio!",model="s2.1-pro-free",)save(audio, "output.mp3")
Everything you needto ship production speech
One API for voice creation, expressive generation, and real-time delivery.
Inline emotion and delivery
Direction written as [tag] syntax, anywhere in the text. Unlimited free-form tags, no separate emotion parameter.
Aber
[Sad]
[Excited]
[satisfied]
[angry]Millions of community voices
Browse and ship from the largest open voice library, or clone your own and build on top of it.
Multi-speaker dialogue
Generate a full conversation in one pass. Available on S2.1 Pro.
Aber
Sarah
EthanInstant voice cloning
Capture timbre and speaking style from a short reference clip. Works across all 80+ languages with no extra training.
My Cloned VoiceRead complex text naturally
Numbers, currency, dates, and units spoken correctly
Streaming and timestamps
Stream audio with word-level timestamps.
Build the Voice Experience Your Product Needs
Create a reusable voice, direct how it performs, and stream it into real-time products.
Clone Any Voice Instantly
with Fish Audio
The Fish Audio AI voice cloning API turns a short reference clip into a reusable voice. The clone captures timbre and speaking style and carries across all 80+ languages with no extra training.
# pip install fish-audio-sdk# Clone from reference audiofrom fishaudio import FishAudiofrom fishaudio.types import ReferenceAudiofrom fishaudio.utils import saveclient = FishAudio() # reads FISH_API_KEYvoice = ReferenceAudio(audio=open("reference.wav", "rb").read(),text="The words spoken in reference.wav.",)# Synthesize with the cloned voiceaudio = client.tts.convert(text="This is my cloned voice.",references=[voice],model="s2.1-pro-free",)save(audio, "clone.mp3")
Fish Audio TTS APIBenchmarks and Specs
One API. Switch models with a single header. Both API models bill at the same flat rate.
POST https://fish.keatska.top/apifish/v1/tts
REAL-TIME FACTOR
Short clip
VOICE CLONING
Native
MULTI-SPEAKER
REST + WebSocket
STREAMING
Word-level
TIMESTAMPS
SAMPLE RATE
CONCURRENT REQUESTS
MP3 · WAV · Opus
AUDIO FORMATS
Flexible pricing for your needs
Frequently asked questions
Get to production this weekend
Get an API key, install the SDK, and ship. No card required.