s-peach-tts 1.0.1

Give your AI a voice. Hear your code finish.

s-peach

Give your AI a voice.

Stop reading logs. Hear when your code is done.

---

s-peach is a TTS notification server for AI agents with support for summaries.

Hook it up to notification services through CLI, REST, or MCP.

Features

• Multiple TTS models — runs on everything from a potato to a high-end rig
• Audio queue — notifications queue up and play in order
• Full featured CLI — use easily from any script
• Daemon mode — runs as a background process; survives terminal closes
• MCP tools — via SSE endpoint at /mcp for agent use
• Doctor diagnostics — s-peach doctor helps you get started
• Voice discovery — audition voices easily s-peach discover
• Voice cloning — Chatterbox supports zero-shot voice cloning
• Security first — IP whitelist and automatic API key generation
• Claude Code hooks — one command to hook you up

Platform Support

macOS and Linux — full support.
WSL — full support (it's Linux).
Windows — partial: s-peach serve, CLI client commands, and Claude Code hooks work.
Daemon management (start/stop/restart) and OS service install require POSIX.

TTS Models

Model	Download	Est. RAM	Voices	Voice Cloning	Speed Control	Notes
kitten-nano (int8)	26 MB	~40 MB	8	No	Yes	Fastest
kitten-micro	43 MB	~80 MB	8	No	Yes	Sweet spot for low-RAM setups
kitten-mini	78 MB	~150 MB	8	No	Yes	Good balance for CPU
Kokoro-82M	339 MB	~300 MB	54	No	Yes	High quality, 9 languages
Chatterbox	3.0 GB	~3.5 GB	Unlimited	Yes	No	Best quality & Zero-shot voice cloning
Chatterbox Turbo	3.8 GB	~4 GB	Unlimited	Yes	No	Faster variant
Chatterbox Multi	~3.0 GB	~3.5 GB	Unlimited	Yes	No	23 languages & voice cloning

Chatterbox models are not installed by default — they add ~750 MB of Python dependencies plus multi-GB model downloads on first use.

Installation

Install with uv

# default install
uv tool install s-peach-tts

# with chatterbox voice cloning (chatterbox requires some pin overrides)
uv tool install "s-peach-tts[chatterbox]" \
  --overrides <(echo -e "numpy>=2.0\ntorch>=2.6.0\ntorchaudio>=2.6.0")

Quick Start

Recommended first run

s-peach init              # scaffold config home dir
s-peach serve             # start the server in foreground
s-peach say "Hello world" # test it out

Discover voices

s-peach voices                                     # see all voices
s-peach discover --model "kitten-micro"            # audition all voices for a model
s-peach discover --model "kokoro" --voices "Onyx,River,Heart" # audition selected voices

Configure

Default settings enable Kokoro 82M. Edit server settings to enable other models. They will download on server startup.

s-peach config             # see help screen for paths and info
s-peach config server      # edit server configurations, enable models etc
s-peach config client      # edit config for default model/voice
s-peach reload             # reload with new option

Summaries

The client can summarize agent output before speaking it aloud. It's using claude -p by default.
Disable or change to something appropriate (ask your AI for help):

summary:
  enabled: true
  command: 'ollama run llama3 "$1"'
  source: ".choices[0].message.content"

Summarize long text before speaking — useful for piping logs, diffs, or agent output:

s-peach say --summary "$(git diff)"
cat build.log | s-peach say --summary
echo "Long explanation..." | s-peach say --summary

The summarization prompts are configurable in the client config.

Run as a service

s-peach install-service     # install as system service
s-peach status              # check that it's running
s-peach say "Hello World"   # test it out
s-peach uninstall-service   # remove the service

Three Ways to Use

1. CLI — send a notification from anywhere:

# Default client configured model and voice
s-peach say "Build complete"
# Select model and voice
s-peach say "Merry Christmas" --model "kokoro" --voice "Santa_US"
# Read command outputs through pipe
echo "Done" | s-peach say --model "kitten-micro" --voice Rosie
# Multilingual (kokoro or chatterbox-multi)
s-peach say "Bonjour le monde" --lang fr --model kokoro
s-peach say "Hola mundo" --lang es --model chatterbox-multi
# Save audio as WAV
s-peach say "Done" --save
# Repeat last message
s-peach say-that-again
s-peach say-that-again --save
# List available voices
s-peach voices

Some flags for s-peach say:

Flag	Description
--help	Display all flags
--model "name"	Select enabled model
--voice "key"	Select voice for model
--speed 1.0	0.1 – 5.0 (kitten/kokoro)
--exaggeration 0.5	0.0 – 2.0 (chatterbox full)
--cfg-weight 0.5	0.0 – 2.0 (chatterbox full)
--lang "en"	Language code (kokoro: en, gb, ja, zh, es, fr, hi, it, pt; chatterbox-multi: 23 languages)
--summary	Summarize the text before speaking
--save	Save audio as WAV to ~/.config/s-peach/output/

2. Claude Code hook — automatic TTS on task completion:

s-peach install-hook claude-code

or add this stop hook to your settings.json

"hooks": {
    "Stop": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "bash ~/.claude/scripts/s-peach-notifier.sh",
            "async": true
          }
        ]
      }
    ]
  }

This wires Claude Code to call s-peach whenever a task finishes. You'll hear the result without looking at the terminal.

Even though summary is on by default through client config, the hook accepts all s-peach say flags so you can can have separate voices for different repos. Sweet!

"hooks": {
    "Stop": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "bash ~/.claude/scripts/s-peach-notifier.sh --model kitten-micro --voice Rosie",
            "async": true
          }
        ]
      }
    ]
  }

3. MCP — tool-based TTS for agents:

Connect your MCP client to http://localhost:7777/mcp (SSE transport).
Available tools: speak, speak_sync, list_voices, say_that_again.

Chatterbox and 0-shot voices

• Put voice samples longer than 5s in ~/.config/s-peach/voices
• Update server config s-peach config server
  • Enable chatterbox or chatterbox-turbo under enabled_models
  • Create a new voice in voices.chatterbox.<voice-name>
  • Save and exit
  • If edited in other editor, reload server with s-peach reload
• Use the voice like usual:
  • s-peach say "Hello World" --model "chatterbox" --voice "Example1"
  • s-peach say "Hello World" --model "chatterbox-turbo" --voice "Example2"

Default voice Bea taken from OpenSLR EmoV_DB

Chatterbox & Chatterbox-multi

Supports creative control:

• Add drama with --exaggeration, default 0.5, range 0.0 - 2.0
  • Higher means more drama.
• Add freedom with --cfg-weight, default 0.5, range 0.0 - 2.0.
  • Higher means more like reference audio.

Chatterbox-turbo

• Faster generation
• Supports paralinguistic tags in text:
  • [laugh], [chuckle], [sigh], [gasp], [sniff], [clear throat], [cough], [groan], [shush] and [pause]

Chatterbox-multi (multilingual)

Supported Languages (ISO 639-1)

• Northern & Western Europe
  English (en), Swedish (sv), Danish (da), Norwegian (no), Finnish (fi), German (de), French (fr), Dutch (nl)

• Southern & Eastern Europe
  Spanish (es), Italian (it), Portuguese (pt), Polish (pl), Russian (ru), Greek (el)

• Asia
  Chinese (Mandarin) (zh), Japanese (ja), Korean (ko), Hindi (hi), Malay (ms)

• Middle East & Africa
  Arabic (ar), Hebrew (he), Turkish (tr), Swahili (sw)

Kokoro

Supported Languages (ISO 639-1)

• Northern & Western Europe
  English (US) (en), English (UK) (gb), French (fr)

• Southern & Eastern Europe
  Spanish (es), Italian (it), Portuguese (pt)

• Asia
  Japanese (ja), Chinese (Mandarin) (zh), Hindi (hi)

• Middle East & Africa
  None currently

Troubleshooting

s-peach doctor        # diagnose everything
s-peach doctor --fix  # auto-fix safe issues

Documentation

• System architecture — tech stack, request flow, component overview
• Feature docs — detailed docs for every feature
• Architectural decisions — why things are the way they are

Contributing

See CONTRIBUTING.md.

License

MIT — see LICENSE.