Tether releases open-source on-device speech model for smart glasses and robots
Tether AI Research released QVAC TranscribePsy, an open-source 254-million-parameter speech recognition model that runs on phones, smart glasses and humanoid robots without cloud access.
Why it matters
The open-source release gives developers speech recognition that operates without an internet connection, aimed at wearable and robot hardware where an always-on microphone and intermittent connectivity are the norm. The accuracy and speed figures come from Tether's own tests, so how the model compares with commercial systems in production use is not established.
Where the sources disagree
Independent verification of the benchmark claims
Benchmarks are company-reported and not independently verified
Blockchain.News
Faster than several competing systems in Tether's tests
Crypto.News
Blockchain.News published the word error rate figures but flagged the claim of leading the parameter class as coming from Tether's own announcement, while Crypto.News said the model performed faster than several competing systems in Tether's tests and did not repeat that caveat.
Key facts
Tether AI Research said on 14 October 2026 that it had released QVAC TranscribePsy, an open-source speech recognition model that runs on the device instead of sending audio to cloud servers.
Reported by Blockchain.News
TranscribePsy is a 254 million parameter multilingual automatic speech recognition model.
254,000,000 parametersReported by Blockchain.News
Tether said the model's Core ML build transcribes ten minutes of audio in 1.65 seconds on an iPhone 17 Pro Max.
1.65 secondsReported by Blockchain.News
Tether reported word error rates of 8.02% on the Open ASR Leaderboard, 6.58% on Multilingual LibriSpeech and 21.56% on Common Voice, with the model leading all tested models in its parameter class.
8.02 %Reported by Blockchain.News
Tether said the Core ML build uses less than half the disk space and 1.7 times less memory than a tested Core ML competitor.
1.7 xReported by Blockchain.News
Tether said TranscribePsy can process audio from smart glasses connected via Bluetooth or Wi-Fi and from a humanoid robot's onboard processor without cloud upload.
Reported by Blockchain.News
The model was released as part of Tether's QVAC AI project, which the company describes as running AI applications directly on consumer devices while keeping user data private.
Reported by Blockchain.News
What happened
Tether AI Research said on 14 October 2026 that it had released QVAC TranscribePsy, an open-source speech recognition model that runs on the device itself rather than sending audio to cloud servers, aimed at smartphones, smart glasses and humanoid robots. The model has 254 million parameters and processes audio locally, including from smart glasses connected over Bluetooth or Wi-Fi and from a humanoid robot's onboard processor; Tether said local processing cuts latency and keeps conversations that may contain personal, workplace or medical details off external servers. Tether said the Core ML build transcribes ten minutes of audio in 1.65 seconds on an iPhone 17 Pro Max while using less than half the disk space and 1.7 times less memory than a tested Core ML competitor. Blockchain.News reported word error rates of 8.02% on the Open ASR Leaderboard, 6.58% on Multilingual LibriSpeech and 21.56% on Common Voice, and noted that Tether's claim of leading all tested models in its parameter class rests on the company's own announcement. Crypto.News reported that TranscribePsy covers 25 European languages on laptops and eight on phones, with dedicated builds optimized for Apple's Neural Engine, and that it sits inside Tether's QVAC project for running AI applications locally. Neither outlet reports independent verification of the benchmark or speed results.
How the story developed
- Official statement
Tether AI Research announces the TranscribePsy release
Tether AI Research said on 14 October 2026 that it had published QVAC TranscribePsy as an open-source, on-device speech recognition model.
- First report
First report by Blockchain.News
Tether releases on-device speech model for smart glasses and robots
- Update
Blockchain.News reports the launch with benchmark figures
The first report gave the parameter count, the three word error rate results and the Core ML speed claim, and noted the results come from Tether's own announcement.
- Independent corroboration
Independently corroborated by Crypto.News
Tether launches offline AI speech model for smart glasses and robots
- Update
Crypto.News adds language coverage and Neural Engine detail
The follow-up gave the 25-language laptop and eight-language phone split, the optimized Apple Neural Engine builds and placed the model inside Tether's QVAC project.
How coverage built up
Independent sources over time, counted the way the consensus panel counts them: a republication of a wire story does not move the line.
- 14 Oct 2026, 14:05 UTC: Blockchain.News — 1 independent source, 1 reports
- 15 Oct 2026, 08:29 UTC: Crypto.News — 2 independent sources, 2 reports
Affected entities
Original sources
| Publisher | Report | Role | Published |
|---|---|---|---|
| Blockchain.News Crypto media | Tether releases on-device speech model for smart glasses and robots | Original | |
| Crypto.News Crypto media | Tether launches offline AI speech model for smart glasses and robots | Independent |