Skip to content

ASR experiment using Google's Universal Sentence Encoder

License

Notifications You must be signed in to change notification settings

daveshap/keras_asr

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Data Sources

  1. LibriSpeech (raw audio and text transcripts)
  2. Universally encoded sentences

Data Prep

  1. Generate MFCC samples from FLAC audio
  2. Generate encoded sentences from transcripts

Datasets Produced

  • MFCC samples
  • Text sentences
  • Encoded sentences

Encoder Model

Train model on:

  • Input: MFCC
  • Output: Encoded sentence

Decoder Model

Train model on:

  • Input: Encoded sentence
  • Output: Text sentence

Inference

  1. Generate MFCC from raw audio
  2. Generate encoded sentence by feeding MFCC to encoder model
  3. Generate text sentence by feeding encoded sentence to decoder model