Tether AI Research releases on-device speech recognition model TranscribePsy for smart glasses and humanoid robots

Article

BlockBeats news,October 14，Tether AI Research has released the open-source multilingual automatic speech recognition model QVAC TranscribePsy, with approximately 254000000, which can run entirely on-device without uploading audio to the cloud. According to the official statement, it can process audio connected to smart glasses via Bluetooth or Wi-Fi in near real time, and can also run on onboard processors in humanoid robots; on the iPhone 17 Pro Max Core ML version, transcribing 10 minutes of audio takes approximately 1.65 seconds.

According to its disclosed evaluation data, TranscribePsy's word error rate (WER) on the Open ASR Leaderboard is WER, 8.02%on Multilingual LibriSpeech, and 6.58%on Common Voice; across tested phones, it was the fastest in its category, approximately 21.56%times faster than the second-fastest model. The model supports 2.6 European languages on laptops and 25 European languages on mobile phones, and releases its model weights and build versions such as 8 under the Apache 2.0 license. Tether GGUF、Core ML Paolo Ardoino stated that he hopes advanced speech CEO can run locally on users' own devices, so data does not need to leave the device.

Original link AI https://m.theblockbeats.info/flash/371796

Key points

Tether AI Research releases open-source multilingual automatic speech recognition model QVAC TranscribePsy。
The model has approximately 254000000, and can run entirely on-device without uploading audio to the cloud.
According to the official statement, the model can process audio connected to smart glasses via Bluetooth or Wi-Fi and can also run on onboard processors in humanoid robots.
On the iPhone 17 Pro Max Core ML version, transcribing 10 minutes of audio takes approximately 1.65 seconds.
The model's word error rate on the Open ASR Leaderboard is 8.02%, 6.58%on Multilingual LibriSpeech, and 21.56%。
The model supports 25 European languages on laptops and 8 European languages on mobile phones.
The model weights and build versions such as GGUF、Core ML are released under the Apache 2.0 license.
Tether CEO Paolo Ardoino stated that he hopes advanced speech AI can run locally on users' own devices, so data does not need to leave the device.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

This is a model that runs speech recognition locally on phones, smart glasses, or robots. Audio does not need to be uploaded to the cloud; the materials also list its speed, error rates, and supported languages on certain devices and test sets.

Why it matters to readers

It demonstrates one application direction for on-device speech recognition models on smart glasses, humanoid robots, and mobile devices.

On-device operation primarily relies on the device itself to process data, rather than sending audio to the cloud. A lower word error rate generally indicates fewer word errors in transcription results.

Risks and unknowns

The materials do not provide an independent official model repository or technical documentation to verify the parameters, evaluation data, and license information.
The materials do not specify actual resource usage, latency, or accuracy across different devices.
The materials do not confirm that smart glasses, humanoid robots, or other manufacturers have integrated the model.

Tether AI Research releases on-device speech recognition model TranscribePsy for smart glasses and humanoid robots

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center