pi-tts

Text-to-speech for the pi coding agent: speaks the final assistant answer once the agent settles, and can be driven manually via /audio.

• Edge backend — Microsoft cloud neural voices via edge-tts (needs network + Python)
• Native backend — fully offline: Windows WinRT (OneCore voices, e.g. cs-CZ Jakub) with SAPI5 fallback; Linux espeak-ng/espeak (voice, rate and pitch derived from the configured voice's locale) with a spd-say streaming fallback
• Vader mode — ffmpeg filter chain (EQ + flanger + echo + rubberband pitch) tuned for a Darth Vader delivery; works on edge and on any native backend that can render a WAV (WinRT/SAPI5/espeak); falls back to plain audio without ffmpeg

Requirements

• Python 3 on PATH (python on Windows, python3 or python elsewhere)
• pip install -r requirements.txt (only for the edge backend)
• ffmpeg + ffplay — optional, needed for the Vader effect and as the preferred player
• Native backend engines:
  • Windows: PowerShell 5.1 (powershell.exe, preinstalled)
  • Linux: espeak-ng or espeak (recommended — enables voice/rate/pitch control and native Vader), otherwise spd-say as a plain streaming fallback

Install

pi install git:github.com/mastnacek/pi-tts

/audio commands

Command	Effect
/audio on / off	Enable/disable speaking (persisted)
/audio stop	Stop current playback
/audio status	State, backend, voice, rate, vader, last error
/audio voice <name>	Set voice (e.g. cs-CZ-AntoninNeural); no arg shows current
/audio backend edge|native	Cloud neural vs offline engine
/audio vader on|off	Toggle the Vader effect
/audio vader classic|vader2|vader3|c3po	Switch between Classic, Vader2, Vader3, and C-3PO profiles
/audio vader2 on|off	Quick shortcut to enable/disable Vader2 mode
/audio vader3 on|off	Quick shortcut to enable/disable Vader3 mode
/audio c3po on|off	Quick shortcut to enable/disable C-3PO droid mode
/audio vader depth <semitones>	Extra pitch shift; negative = deeper; auto = 0 on edge, −3 on native
/audio rate ±N%	Speech rate (e.g. +10%, -10%)
/audio say <text>	Speak text immediately

A new user prompt interrupts playback. Config persists in ~/.pi/agent/pi-tts.json.

Environment variables

Variable	Default	Purpose
PI_TTS_VOICE	cs-CZ-AntoninNeural	Default voice
PI_TTS_RATE	+0%	Default rate
PI_TTS_PITCH	+0Hz	Default pitch
PI_TTS_MAXLEN	1500	Max characters spoken
PI_TTS_BACKEND	edge	Default backend
PI_TTS_VADER	off	Start with Vader on
PI_TTS_VADER_PROFILE	classic	Vader profile (classic, vader2, vader3, c3po)
PI_TTS_VADER_PITCH	-30Hz	Vader base pitch
PI_TTS_VADER_RATE	-30%	Vader base rate
PI_TTS_VADER_VOLUME	61.5	Classic Vader make-up gain
PI_TTS_VADER2_VOLUME	2.4	Vader2 make-up gain
PI_TTS_C3PO_VOLUME	2.2	C-3PO make-up gain
PI_TTS_VADER_DEPTH	0 edge / -3 native	Extra semitone shift
PI_TTS_NATIVE_API	auto	winrt / sapi / auto
PI_TTS_CA_BUNDLE	—	Extra CA bundle for TLS-intercepting antivirus proxies

speak.py standalone

python speak.py --voice cs-CZ-VlastaNeural --rate -10% "Text"
python speak.py --file response.txt --backend native --vader --depth -5

Markdown is stripped (code blocks become "code omitted"), text is truncated at PI_TTS_MAXLEN.

License

MIT