More-inclusive speech recognition with cross-utterance rescoring
Top 3% at ICASSP. Amazon's new speech recognition breakthrough tackles the bias problem that's been plaguing voice AI.

Why it matters
Amazon's research addresses a critical AI fairness issue that affects real business applications - speech recognition systems that fail on underrepresented pronunciations, potentially excluding entire user groups from voice-enabled products and services.
The key facts
4 to knowTop 3% paper at ICASSP conference
Graph-based label propagation adaptation
Targets underrepresented pronunciations
Cross-utterance rescoring technique
Go to the source
Amazon Scienceamazon.science
Publisher excerpt: In a top-3% paper at ICASSP, Amazon researchers adapt graph-based label propagation to improve speech recognition on underrepresented pronunciations.


