FlexSR Limited

The novel FlexSR (WIPO Patent Granted), is flexible speech recognition accommodating any combinations of languages, different speaker accents and pronunciations, without acoustic or end-to-end models, enabling ASR on an edge device, mobile or chip built into a wearable, for multi-language multi-accent, hands-free, deployment in operational environments. This is a massive system differentiator that enables greater than a ten fold reduction in software development costs and time to market.

FlexSR evolved from University of Oxford, Linguistic Professors, Aditi Lahiri and Henning Reetz, researching how the human brain identifies and matches linguistic features from speech, to the words in its mental lexicon. It is incredibly fast and efficient on processing compared to the computer modelling approach and the closest AI methodology to human brain recognition. The POC demonstrable code was funded by ERC.

It requires no specialist data modelling skills as it can be applied to any Lexicon conforming to the IPA, enabling entire catalogues of Lexicons, Languages, Accents to be dropped into the system whilst in the field of operation. This enables immediate adaption to different pronunciations e.g. Scottish, Geordie, Mancunian, Brummie, Irish, Indian. The system notation required for these variants is small and needs no model training.