text-to-speech

name	text-to-speech
description	Generate speech audio from text. Use when a user asks to generate speech, read text aloud, create audio from text, or do text-to-speech / TTS.

Text-to-Speech

Generate natural-sounding speech from text using Kokoro TTS running locally on CPU. No external API keys required.

Bundled Scripts

tts.py — Generate speech from text

# Text from argument (default voice: af_heart, default format: ogg)
uv run scripts/tts.py "Hello, world!" -o output.ogg

# Text from stdin
echo "Hello world" | uv run scripts/tts.py - -o output.ogg

# Text from file
uv run scripts/tts.py --file input.txt -o output.ogg

# Different voice
uv run scripts/tts.py "Hello!" -o output.ogg --voice am_adam

# British English voice
uv run scripts/tts.py "Good day!" -o output.ogg --voice bf_emma

# WAV output (larger files)
uv run scripts/tts.py "Hello!" -o output.wav

# Adjust speed (0.5 = slow, 2.0 = fast)
uv run scripts/tts.py "Hello!" -o output.ogg --speed 1.2

# List all available voices
uv run scripts/tts.py --list-voices

Setup

Dependencies (kokoro-onnx, soundfile) are declared inline via PEP 723 and handled automatically by uv run. Model files (~336 MB total) are downloaded on first use and cached in ~/.cache/kokoro-tts/.

First run is slower due to dependency resolution and model download. Subsequent runs take a few seconds for short texts.

Voices

Default voice is af_heart (US English, female). Use --list-voices to see all 30+ voices across English (US/GB), French, Italian, Japanese, and Mandarin. Language is auto-detected from the voice name prefix; override with --lang if needed.

Output Formats

• OGG Vorbis (default) — compact, good quality for voice (~6x smaller than WAV)
• WAV — uncompressed, use when lossless quality is needed

Format is auto-detected from the output file extension, or can be set explicitly with --format.

Notes

• Runs entirely on CPU — no GPU required.
• OGG Vorbis at 24 kHz mono produces very compact files suitable for sharing in chat.
• For long texts, generation may take proportionally longer. Short messages (a few sentences) take just a few seconds.