PyDecoder

Basic Speech Recognition Decoder implemented in Python.

Requirements

• Python 3.7
• Numpy

Not tested in versions <3.7, but it should work as well.

Structure

• code: Python code
• models: Models required to do the recognition
  • Gaussian Mixture Acoustic Model (plain and binary)
  • Search Graph (plain and binary)
• samples: Example sample to recognise, with:
  • WAV file
  • MFCC acoustic features (plain and binary)
  • Text file with the transcription (plain and binary)

Run the code

To recognise the provided sample, run:

cd code
python -m decoder.Decoder

The output will be generated in the deco.info.log file.

TO DO

• Include rescoring LM.
• Generate Wordgraphs.
• Advanced pruning.