Gradium's 31M-parameter TTS model runs 2.6x faster than a 2.37B model in hands-on test
10/23/2026 — 10/23, 07:18·2 sources (all)·2 reports (all)
Key takeaways
- Gradium has open-sourced Phonon, a family of on-device text-to-speech models;
- the English Phonon 31m variant ships under Apache 2.0 at 64.5 MB and starts speaking in about 7 ms on a laptop CPU. In a hands-on Mac mini t…
Background & analysis
On October 23, 2026, MarkTechPost reported that Gradium has open-sourced Phonon, a family of text-to-speech models designed to run locally on devices. The English release, Phonon 31m, is distributed under the Apache 2.0 license and fits in 64.5 MB. According to the report, it begins speaking in about 7 ms on a laptop CPU. The model has 31M parameters. The material does not include comparisons with other models, nor does it mention later updates or additional versions, so the public record currently stops at the open-source release.
AI-generated from 1 reports · updated 6 hours ago
Latest turnThe author swapped the local TTS setup on a Mac mini and compared two models on wall-clock time and Whisper round-trip word error rate. A 31.2M-parameter model ran 2.6x faster than the previous 2.37B model, with prose coming back at 0.0% WER. The takeaway is that model size is not a reliable predictor for choosing a local TTS model.
- Heat index
- 177
- All sources
- 2
- All reports
- 2
- First seen
- 7 hours ago
Reports on this story headlines open the original
The author swapped the local TTS setup on a Mac mini and compared two models on wall-clock time and Whisper round-trip word error rate. A 31.2M-parameter model ran 2.6x faster than the previous 2.37B model, with prose coming back at 0.0% WER. The takeaway is that model size is not a reliable predictor for choosing a local TTS model.
DEV Community · AIAI score 55Gradium has open-sourced Phonon, a family of on-device text-to-speech models, MarkTechPost reports. The English Phonon 31m ships under Apache 2.0, fits in 64.5 MB and starts speaking in about 7 ms on a laptop CPU. The release targets fully local inference; no head-to-head benchmark results were included.
MarkTechPostAI score 53
Other stories people are talking about
- Heat index 1279PHP ships emergency patch for two critical AI-found flaws22 sources
- Heat index 935RisingMeta Creates Non-Voting Class C Shares to Fund AI Spending14 sources
- Heat index 805Anthropic launches Claude Duet agent in research preview14 sources
- Heat index 583DeepSeek closes roughly RMB 100B round with CATL and Tencent backing10 sources
- Heat index 553METR finds flaws in OpenAI's agent monitoring system12 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion