MIT OR Apache-2.0freeRuns locallyVideo & Audio

Phonon: On-Device Open TTS (Apache-2.0 31M)

On-device TTS from Gradium. Default 31M English weights are Apache-2.0; 40M variants are CC BY-NC.

Official Phonon hero art from the Gradium xn-ptts README — Official project image. Source · License: MIT OR Apache-2.0.

License: MIT OR Apache-2.0
Stars: 209 (2026-10-22)
Platforms: Linux, macOS, Windows, iOS, Android, Browser, CPU
First seen: 2026-05-03

TL;DR. Phonon is On-device TTS from Gradium. Default 31M English weights are Apache-2.0; 40M variants are CC BY-NC. License: MIT OR Apache-2.0 (verified). Platforms: Linux, macOS, Windows, iOS, Android, Browser, CPU. Price: free. Runs locally: yes. It is an alternative to ElevenLabs.

Name	Phonon
Language	Rust
License	MIT OR Apache-2.0
Pricing	free
Platforms	Linux, macOS, Windows, iOS, Android, Browser, CPU
Runs locally	Yes
Stats	209 stars · 19 forks (as of 2026-10-22)
First seen	2026-05-03
Repository	https://github.com/gradium-ai/xn-ptts

Phonon is Gradium’s on-device text-to-speech stack: a Rust runtime (ptts / gradium-ai/xn-ptts) plus small checkpoints on Hugging Face. Official hosts we opened on 2026-10-22: gradium/phonon-31m-en, gradium/phonon-40m-en, gradium/phonon-40m-multilingual, the runtime LICENSE-MIT and LICENSE-APACHE, release v0.4.0 (published 2026-10-20T13:36:55Z), and the product page gradium.ai/phonon.

GitHub API that day: 209 stars, 19 forks, created 2026-05-03T11:30:57Z, last push 2026-10-21T13:38:23Z, language Rust. GitHub’s license field reports Apache-2.0; the README badge and the two LICENSE files are MIT OR Apache-2.0. We follow the files.

This page hangs off open ElevenLabs alternatives next to Drama 3 and GPT-SoVITS. That hub names a product class. We do not rank voice quality against ElevenLabs.

License split (read this first)

Artifact	Host	License (this check)	Commercial
Runtime `ptts` / xn-ptts	github.com/gradium-ai/xn-ptts	MIT OR Apache-2.0	Permissive code license. Not legal advice.
phonon-31m-en (default)	huggingface.co/gradium/phonon-31m-en	Apache-2.0, ungated	Card: weights, tokenizer, and voices in this repo allow commercial use.
phonon-40m-en / fr / de / es / pt	huggingface.co/gradium/phonon-40m-*	CC BY-NC 4.0, gated: auto	Non-commercial unless you have a Gradium agreement.
phonon-40m-multilingual	huggingface.co/gradium/phonon-40m-multilingual	CC BY-NC 4.0, gated: auto	Same NC rule. Product page: “For commercial use, talk to sales.”
Voices	named `gradium/phonon-voices` on the v0.4.0 notes	Own terms	Do not assume the 31m Apache grant covers every catalog voice.

Do not treat the 40m line as “Apache / commercial OK.” The 31m English checkpoint is the permissive-weight row.

What we verified (2026-10-22)

Field	31m-en	40m-en	40m-multilingual	Source
Created	2026-10-19T12:02:53Z	2026-10-19T12:01:47Z	2026-10-19T12:02:43Z	HF API
Gated	false	auto	auto	same
Likes / downloads	67 / 3,590	53 / 562	33 / 349	same
License tag	apache-2.0	cc-by-nc-4.0	cc-by-nc-4.0	HF `cardData`
Params (card)	31M excl. tokenizer	40M	40M	READMEs
q8 weights (card)	64.5 MB	74.6 MB	74.6 MB	same
Languages	English	English	en, fr, de, es, pt	same
Output	24 kHz mono	same family	same family	31m card

The 31m card describes a flow-matching transformer that predicts Mimi-codec latents one 80 ms frame at a time. Phonon “builds on Pocket TTS and the Mimi codec, developed by Kyutai.” Those are copied credits.

How to run (from the card / v0.4.0, not tested)

Default path downloads phonon-31m-en:

uvx ptts --lang en --quant q8 \
  --voice Easton "Hello from Phonon." -o speech.wav

Also documented: pip install ptts; brew install gradium-ai/tap/ptts; cargo install ptts --locked --features cli; npm install phonon-tts; Docker ghcr.io/gradium-ai/ptts-openai-server:0.4.0 (OpenAI-compatible speech API + Wyoming). 40m checkpoints need hf://gradium/phonon-40m-…, an accepted gate, and HF_TOKEN. forkalt has not run these commands.

v0.4.0 is labeled “the first stable release.” The runtime is Rust and “needs no PyTorch.”

ElevenLabs alternatives — Drama 3 (35B MoE, Community License) and GPT-SoVITS (MIT few-shot cloning).
Transcription (the other direction): whisper.cpp on Descript alternatives.
Speaker turns, not TTS: pyannote community-1 on Otter.ai alternatives.

Still images and posters that sit next to a voiceover are a pixel-editor job, not TTS: PhotoCraft vs GIMP and the docs-based PhotoCraft install outline.

FAQ

Can I use Phonon commercially?

Split the artifacts. The ptts runtime is MIT OR Apache-2.0. Default English weights gradium/phonon-31m-en are Apache-2.0; the card says those weights, tokenizer, and voices in that repo allow commercial use. Every 40m checkpoint we opened (en/fr/de/es/pt and multilingual) is CC BY-NC 4.0 and gated: auto. Commercial use of 40m weights needs a Gradium agreement. Voices have their own terms. Not legal advice.

Is Phonon an ElevenLabs alternative?

It is local on-device TTS (Rust runtime, no PyTorch). ElevenLabs is a hosted voice product. We have no sourced quality bake-off. See ElevenLabs alternatives.

How big is the default model?

phonon-31m-en is 31M parameters excluding the audio tokenizer. The card lists model.q8.gguf at 64.5 MB, English only, 24 kHz mono. Hugging Face likes / downloads in the 2026-10-22 API check: 67 / 3,590. Official hardware claim is laptop CPU, phone, or a browser tab. Untested here.

How do I run it?

The card and v0.4.0 notes show `uvx ptts --lang en --quant q8 --voice Easton "Hello from Phonon." -o speech.wav`, `pip install ptts`, `cargo install ptts --locked --features cli`, Homebrew `gradium-ai/tap/ptts`, and `npm install phonon-tts`. First run downloads Phonon 31m. forkalt has not run these commands.

Sources & verification

Last verified 2026-10-22 · Last updated 2026-10-22· Edited by forkalt editors · Editorial policy

License notes are not legal advice. Read the original LICENSE. Trademark & disclaimer.