hacker.org

Basically, the method consists of a predefined Huffman tree and a non-ambiguous context free grammar in Greibach normal form. Take the following production: noun -> Johnny | Drake When parsing a text, whenever Johnny is found in place of a "noun", a 0 is noted (first choice); whenever Drak...

Wow. This was a rollercoaster. I used the SpeechRecognition library for Python to predict all 22,536 snippets (Google Speech Recognition). The results were ok but quite a lot of bytes were wrong. (Padding all snippets with a second of silence at the beginning greatly improved the results, however.) ...