Python Deep Learning - Second Edition

Speech recognition pipeline

Speech recognition tries to find a transcription of the most probable word sequence considering the acoustic observations provided:

transcription = argmax(P(words | audio features))

This probability function is typically modeled in different parts (note that the normalizing term P (audio features) is usually ignored):

P (words | audio features) = P (audio features | words) * P (words)

= P (audio features | phonemes) * P (phonemes | words) * P (words)

Phonemes are basic units of sound that define the pronunciation of words. For example, the word "bat" is composed of three phonemes: /b/, /ae/, and /t/. Each phoneme is tied to a specific sound. Spoken English consists of around 44 phonemes.

Each of these probability ...