Voice of India: A Large-Scale Benchmark for Real-World Speech Recognition in India
Researchers have introduced "Voice of India," a new benchmark designed to improve automatic speech recognition (ASR) for 15 major Indian languages. Unlike previous benchmarks that used scripted speech, this dataset comprises unscripted telephonic conversations from over 36,000 speakers, totaling 536 hours. The benchmark accounts for spelling variations common in Indian languages and analyzes ASR performance geographically, revealing disparities across regions and factors like audio quality and device type. AI
IMPACT Addresses limitations in current ASR systems for Indian languages, potentially improving accessibility and usability of voice technologies across diverse regions.