← Back to the collection

Text to Speech

Speak text aloud, or save it as audio, with local models. Ask in plain words (“read this in a cheery female British voice, a bit faster”) and the skill picks the voice and speed.

ollama-talk: Ollama’s Phonon model. It matches a description to a voice using the Gradium voice catalog, which covers locale, gender, age and a prose description for each voice. It only offers voices the installed tag actually bundles. Speed is applied after generation with ffmpeg’s pitch-preserving atempo, because Phonon has no speed control of its own.

For the reverse direction (audio to text), see the work-with-media plugin’s macwhisper-cli skill.

Installation

Run these in a terminal (not inside a Claude Code session):

claude plugin marketplace add jtsternberg/claude-plugins
claude plugin install text-to-speech@jtsternberg

Prerequisites

Ollama 0.40.9 or later, running.
ollama pull phonon. This is the 31M English model (Apache 2.0) and the one with named voices; phonon.py voices lists them. The :de, :es, :fr, :pt, :multilingual and :40m-en tags are 40M models under CC BY-NC, so non-commercial use only. phonon:en carries only the default voice.
ffmpeg, needed only for speed changes and MP3/M4A output: brew install ffmpeg.
macOS afplay for playback. On other systems, write a file with --out.

Usage

python3 skills/ollama-talk/scripts/phonon.py doctor
python3 skills/ollama-talk/scripts/phonon.py voices --gender female --locale british
python3 skills/ollama-talk/scripts/phonon.py say --voice "Imogen" --speed 1.25 "Hello there."
python3 skills/ollama-talk/scripts/phonon.py say --voice "The Snow Queen" --file story.txt --out story.mp3

Without --out, say plays the audio and keeps a copy at /tmp/phonon-<epoch>-<pid>.wav. Copies older than a day are removed on the next run.

Structure

plugins/text-to-speech/
├── .claude-plugin/plugin.json
├── README.md
└── skills/
    └── ollama-talk/
        ├── SKILL.md
        ├── references/voices.jsonl   ← Gradium voice catalog
        ├── scripts/phonon.py          ← stdlib-only CLI over the Ollama API
        └── tests/test_phonon.py

To refresh the catalog: curl -sL https://huggingface.co/gradium/phonon-voices/resolve/main/voices.jsonl -o skills/ollama-talk/references/voices.jsonl.