The goal was to achieve good word error rate using open source software. After trying some of the existing software available, I've found out that a modified version of Kaldi is able to produce speech recognition results on par with it's proprietary alternatives. This blog post will walk through the steps I've taken in getting there and also how to modify an existing Kaldi model to fit for domain specific words and expressions.
Hello fellow Researcher, This blog came to life to share some of my adventures in hopes that it will help others facing the problems I've encountered. Should you have any questions, critics or suggestions, please let me know in the comments section or contact me using the social links at the bottom.