Google AI researchers working with the ALS Remedy Improvement Institute right now shared particulars about Venture Euphonia, a speech-to-text transcription service for individuals with talking impairments. In addition they say their strategy can enhance automated speech recognition for individuals with non-native English accents as properly.
Folks with amyotrophic lateral sclerosis (ALS) usually have slurred speech, however present AI programs are usually educated on voice knowledge with none affliction or accent.
The brand new strategy is profitable primarily because of the introduction of small quantities of knowledge that represents individuals with accents and ALS.
“We present that 71% of the development comes from solely 5 minutes of coaching knowledge,” in response to a paper revealed on arXiv July 31 titled “Personalizing ASR for Dysarthric and Accented Speech with Restricted Information.”
Personalised fashions had been capable of obtain 62% and 35% relative phrase error price (WER) enchancment for ALS and accents respectively.
The ALS speech knowledge set consists of 36 hours of audio from 67 individuals with ALS, working with the ALS Remedy Improvement Institute.
The non-native English speaker knowledge set known as L2 Arctic and has 20 recordings of utterances that final one hour every.
Venture Euphonia additionally makes use of strategies from Parrotron, an AI instrument for individuals with speech impediments launched in July, in addition to fine-tuning strategies.
Written by 12 coauthors, the work is being offered at Worldwide Speech Communication Affiliation, or Interspeech 2019, which takes place September 15-19 in Graz, Austria.
“This paper’s strategy overcomes knowledge shortage by starting with a base mannequin educated on hundreds of hours of normal speech. It will get round sub-group heterogeneity by coaching personalised fashions,” the paper reads.
The analysis, which a Google AI weblog publish highlighted right now, follows the introduction of Venture Euphonia and different initiatives in Could, resembling Reside Relay, a function to make cellphone calls simpler for deaf individuals, and Venture Diva, an effort to make Google Assistant accessible for nonverbal individuals.
Google is soliciting knowledge from individuals with ALS to enhance its mannequin’s accuracy and is engaged on subsequent steps for Venture Euphonia, resembling utilizing phoneme errors to scale back phrase error charges.