This was made using the Web Speech API and a matching system I came up with that uses Levenshtein distances and tokenized string matching. I wrote up some demo code here: https://github.com/benwasser/WebSpeechAPI