- Who it helps
- Developers building spoken interfaces, read-along text or captions.
- Availability
- Public Apache2.0 runtime for Gradium’s Phonon 31M English model. The separate 40M language models use a non-commercial license.
Where the timing comes from
The runtime uses the model’s attention and audio loudness to estimate word boundaries while streaming 24 kHz speech. That could keep highlighted text aligned with playback; timing accuracy was not independently measured.
Download and performance limits
The maker reports a 44.3 MB first visit and 115 times real-time synthesis on an M3 Max. Those figures describe its own benchmark, with separate warm-cache and first-download conditions. The base Gradium card lists a different weight-file size because it describes a different packaging format.
Blixt’s Phonon runtime adds word timing to browser speech ↗Original Phonon 31M model card ↗
What you can do
- Inspect the benchmark conditions and test pronunciation and timing on your own device. Open reference ↗
Where the evidence stops
README and original model card checked November 4. No runtime installed, audio heard or browser performance tested. Browser support and timing quality remain author-reported.
Sources read. Primary sources checked November 4, 2026. No third-party software, model inference or complete workflow executed.
Sources & further reading
GitHub · implementation
Blixt’s Phonon runtime adds word timing to browser speech ↗
Identifies the subject, documented capability or specific verification limit.
Blixt · page checked 2026-11-04
21 stars · observed 2026-11-04 19:04 UTC
Hugging Face · official
Original Phonon 31M model card ↗
Identifies the subject, documented capability or specific verification limit.
Gradium · page checked 2026-11-04
Coverage note: Independent attention and end-to-end execution remain unverified.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
Blixt’s WebGPU runtime generates speech in a browser and exposes start and end times for each word.