Phonon

Phonon is a WebGPU runtime for text-to-speech models in the browser that generates speech and word timings.

blixt/phonon is a WebGPU runtime for the Phonon 31M text-to-speech model by Gradium. It runs entirely in the browser, making 24 kHz English speech while streaming the audio and providing start and end times for every word and token. The runtime also supports Gradium larger 40M models in English, French, German, Spanish, and Portuguese. It downloads model weights converted to 8-bit integers and caches them locally in the browser. It falls back to running on the CPU in WebAssembly if WebGPU is unavailable.

Phonon at a glance

Best for: Software engineer, Researcher, Writer
Runs on: Web
Pricing: Free
Released: November 2026
Category: AI audio tool
Website: github.com/blixt/phonon

Key features

Generates 24 kHz English speech
Provides start and end times for every word and token
Streams audio during generation
Supports variable speaking speeds from 0.77x to 1.3x
Runs on WebGPU or CPU via WebAssembly
Downloads and caches 8-bit integer quantized weights

Use it for

Generating text-to-speech audio directly inside web browsers
Syncing text highlights with spoken audio using word timings
Running offline or browser-based speech applications

Alternatives to Phonon

audio.cpp — Runs modern audio models locally with high-performance, native C++ execution.
runs in your browser — It runs locally in your web browser using WebGPU.
kokoro-onnx — Runs local TTS inference with efficient ONNX runtime performance.
Speech-to-Speech — Fully modular, swappable components with local or hosted execution options.
browser-whisper — Transcribes audio and video locally with zero servers involved.
sherpa-onnx — Runs complex audio AI tasks offline on local devices and embedded hardware.
AuK — It handles zero-shot TTS, editing, and enhancement via natural language.
whisper.cpp — It provides high-performance, offline speech recognition on diverse hardware.
VoxCPM — Provides tokenizer-free multilingual speech with controllable voice creativity.
FireRedAudio — A unified model handling understanding and generation with decoupled continuous representations.
Babytalk — Runs entirely on-device with no cloud or internet required.
transcribe.cpp — Runs 16 model families and 60+ variants with GPU and CPU acceleration.

All Phonon alternatives