Utter

Local text-to-speech and speech-to-text web app powered by Kokoro TTS and Whisper, optimized for Apple Silicon.

Everything runs on your machine. No cloud APIs, no data leaves your computer.

Install

Requirements:

• Mac with Apple Silicon (M1/M2/M3/M4)
• macOS 13 Ventura or later
• ~2.5 GB of disk space (Python packages + AI models)

Paste in Terminal:

curl -fsSL https://raw.githubusercontent.com/vishoo7/Utter/main/install.sh | bash

The script installs everything automatically (~5-10 min depending on internet speed): Homebrew, Python, AI models, and a ready-to-use app. When done, launch Utter by:

• Double-clicking Utter in /Applications, or
• Running utter from Terminal

To uninstall: bash ~/.utter/uninstall.sh

Features

Text to Speech — Kokoro-82M via kokoro

• 3 built-in voices (af_heart, am_michael, bf_emma)
• Outputs M4A files (64kbps AAC) — 30 minutes of speech is ~15MB
• Smart filenames based on the first 5 words of input text
• Generation history with playback and download

Speech to Text — Distil-Whisper Large V3 via mlx-whisper

• Drag-and-drop audio file upload
• Runs natively on Apple Silicon via MLX
• Copy transcript or send it directly to the TTS tab

Manual Setup

If you prefer to set things up yourself instead of using the install script:

Requirements: macOS with Apple Silicon, Python 3.10+, espeak-ng

git clone https://github.com/vishoo7/Utter.git
cd Utter
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
brew install espeak-ng
./run.sh

Then open http://localhost:5757. On first run, the Kokoro model (~330 MB) and Whisper model (~750 MB) will download automatically.

Project Structure

Utter/
├── server.py          # Flask server (localhost:5757)
├── tts.py             # Kokoro TTS: text → WAV → M4A
├── stt.py             # Whisper STT: audio → text
├── templates/
│   └── index.html     # Single-page dark mode UI
├── static/            # Generated audio files
├── install.sh         # One-line installer
├── uninstall.sh       # Clean removal
├── launcher.sh        # App launcher (server + browser)
├── run.sh             # Dev launch script
└── requirements.txt

How It Works

• TTS: Text is fed through Kokoro's KPipeline, which generates audio chunks. Chunks are concatenated, written as WAV, then converted to M4A via macOS afconvert.
• STT: Uploaded audio is transcribed using mlx-whisper with the distil-whisper/distil-large-v3 model, running natively on Apple Silicon via the MLX framework.
• Both operations run in background threads so the UI stays responsive, with status polling every second.