AIKrisp releases voice-isolation benchmark with a reported 73% drop in word error rate
A post made in partnership with Krisp says its new benchmark compares original and voice-isolated audio across 265 real recordings and 11 speech-to-text configurations.
TLDR
A post made in partnership with Krisp reports that voice isolation reduced overall word error rate from 23.24% to 6.24% in its benchmark—a 73% reduction. Reported rates fell from 31.84% to 6.30% for workplace recordings and from 23.83% to 6.86% for call-center recordings. The post says Krisp released an open benchmark and dataset, and that people with speech-to-text setups can reproduce the experiments using their own audio to see whether voice isolation helps.
