tim-shea / speechvis

speechvis

This project provides a set of tools for segmenting, analyzing, and visualizing recorded speech. In particular, the goal is to treat segmented speech from the Warlaumont IVFCR corpus as high-dimensional time-series, and provide a library of analyses to reveal the structure within these high-dimensional data.

Features

• Parse LENA ITS files to extract segment boundaries and labels for entire recordings.
• Plot speaker counts, durations, intervals, and volubility for parsed recordings.
• Segment WAV file recordings to save segments as individual, categorized files.
• Generate and plot acoustic analyses for individual segments.
  • Waveform
  • Power Over Time and Power Over Frequency
  • Spectrogram and Formant Frequency Detection
  • Mel-Frequency Filter Bank and Mel-Frequency Cepstral Coefficients

Examples

Parse a recording and display speaker segment counts

Follow the normal process for importing data from the used LENA recorder to your LENA workstation. Use the LENA software export to save the ITS and WAV files.

From a terminal at the root of the speechvis library:

ipython -i speechvis.py

In the python interpreter:

root = <new root location>
ids.append(<new recording id>)
recording = Recording(root, ids[-1])
plot_speaker_counts(recording)

Display a Principal Components decomposition of the MFCCs for a speaker

In the python interpreter:

recording = Recording(root, ids[<Rec #>])
plot_mfcc_pca(recording, 'CHN') # Change CHN to any LENA speaker category