First technical prototype online

January 24, 2021 by Steven

gongbu.ai is online and serves at the moment as a prototype to validate the technical approach.

The current version is not nearly as precise or as performant as I would like it to be in the future, but it hints at what would be technically possible.

The speech recognition model that was used is trained with 100 hours of clean speech from the LibriSpeech corpus. Male and female speakers are equally represented in this corpus. Speakers with accents closer to US English, on the other hand, are disproportionately represented in this corpus. That's why theoretically, speaking with an american english accent and with little interfering noise should be best understood by gongbu.ai.

Feel free to play with the current version and check the updates coming soon!