~90ms to first sound

Talk to your Claude

Ultra-fast local TTS for AI agents. Your agent speaks, you hear it instantly. No cloud. No API keys. No latency.

npx skills add EmZod/Speak-Turbo

Works with Claude Code, Amp, Opencode, Pi, Cursor

$speakturbo "Hello, I'm your agent"
⚡ 92ms▶ playing✓ done

Why Speak Turbo?

Your agents work hard. But they're silent. You're juggling terminals, missing outputs, losing context. Give them a voice.

Instant

~90ms to first sound. Cloud TTS takes 500ms+. This feels like your agent is in the room.

Private

100% local. Your text never leaves your machine. No API keys. No cloud bills.

Agent-native

Built for CLI. Your agent calls speakturbo and speaks.

When you need it

Real scenarios from agent users

“Running 4 Claude sessions. Need to know which one just finished.”

→ Agent speaks: “Session 3 complete”

“Long task running. Don't want to keep checking.”

→ Agent speaks when it needs you

“Want to vibe with my model, not stare at walls of text.”

→ Conversational, multimodal flow

“Don't want to pay ElevenLabs for API calls.”

→ Free forever. Local. No limits.

8 Built-in Voices

High-quality voices, ready to use

albaFemaledefault
Listen
mariusMale
Listen
javertMale
Listen
jeanMale
Listen
fantineFemale
Listen
cosetteFemale
Listen
eponineFemale
Listen
azelmaFemale
Listen

How it works

Lightweight. Fast. Local.

┌─────────────────┐
│   speakturbo    │
│   (Rust, 2.2MB) │
└────────┬────────┘
         │ HTTP :7125
         ▼
┌─────────────────┐
│     daemon      │
│ (Python + MLX)  │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Audio Output   │
└─────────────────┘
2.2MB
Rust CLI
Realtime
100%
Local

Get started

One command. Apple Silicon.

For AI Agents

npx skills add EmZod/Speak-Turbo

CLI only

pip install pocket-tts uvicorn fastapi