1 comment

[ 3.3 ms ] story [ 9.8 ms ] thread
Hello all, We here on the rev.com speech research team noticed that existing tools for text alignment (used mainly to calculate Word Error Rate (WER) for speech recognition systems) tend to run slowly and consume large amounts of system resources for longer texts. We had an internal tool we developed to calculate WER efficiently using FSTs and lazy algorithms, and today we released the tool as open source (Apache 2) for all to use. Myself and other members of our team will be around to answer questions and hear feedback!