Speech Processing

A simple python toolkit that supports multiple analysis tasks on an audio file of speech. This is in service of "Speaker+", a web-based intelligent training environment that auguments close-loop skill acquisition.

import SpeechProcessing as SP
config = {'username':'USERNAME','password':'PASSWORD'}
speech_transcript = SP.Audio(config,'RELATIVE_PATH.wav'）
# return an Audio object list with sentences and corresponding timestamps
phonetic_info = speech_transcript.segment().output()
# get further phonetic information of each sentences and output a json file as shown in 'test_transcript.json'

• Speech Recognition: transcripts in a stence-level under IBM Watson Cloud Service.
• Speech Analysis: speaking rate(syl/s), fillers (fillers/minute), pitch variety(Hz)

Requirements

• Python 2.7(recommended and tested), 3.4, 3.5, 3.6(tested)

• IBM Watson Cloud (for speech recognition)

  • Installation: pip install watson-developer-cloud
  • Credentials: IBM Cloud Dashboard Page
    • usernames are strings of the form XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
    • passwords are mixed-case alphanumeric strings

• NLTK (for syllable dictionary)

  • Installation: pip install nltk

• Parselmouth (for pitch)

  • Installation: pip install praat-parselmouth

Citation

@article{parselmouth,
    author = "Yannick Jadoul and Bill Thompson and Bart de Boer",
    title = "Introducing {P}arselmouth: A {P}ython interface to {P}raat",
    journal = "Journal of Phonetics",
    year = "in press"
}

@misc{praat,
    title = "{P}raat: doing phonetics by computer [{C}omputer program]",
    author = "Paul Boersma and David Weenink",
    howpublished = " Version 6.0.37, retrieved 3 February 2018 \url{http://www.praat.org/}",
    year = "2018"
}