Voice versus Personality

When we hear someone’s voice for the first time, chances are there is already an idea taking shape in our heads about what type of person they are. Research shows there might be some links between the sound of someone’s voice and their personality. We investigated whether extraverted people have a lower voice pitch, a louder voice and whether they read faster than introverted people do.

We collected audio recordings of our participants reading out loud a fairytale and asked them to fill in a questionnaire to estimate their personality type.

If you are interested in our process and results, you can take a look at our presentation or read our report.

This was a course project in “Statistics in Connected Healthcare” at the Hasso-Plattner Institute in 2022.

Code

The audio analysis is done with the script audioAnalyzer.py. Please install all requirements before executing.

For a given audio recording in WAV format lying in the wav_audios/ directory, the script will:

1. generate an amplitude plot and save it under wav_audios/plots/,
2. extract the voiced speech and save it as a new wav file under wav_audios/extracted/,
3. retrieve the speaking fundamental frequency SFF of the speech audio, its duration and the standard deviation of the fundamental frequency f0,
4. generate a spectrogram figure of the fundamental frequency of the speech and save it under wav_audios/plots/,
5. measure the loudness of the complete audio as well as of its voiced speech part in db LUFS,
6. measure the reading duration without the leading and trailing silence in the audio,
7. write all found data to wav_audios/f0.csv as a line.

All data is summarised with the script data-analysis.R. It analyses the answers to the questionnaire regarding the personality type, calculates the extraversion score of someone and classifies them as an introvert or extravert.

Examples

You can find exemplary recordings used in the presentation in examples/.

Data

We will not publish the actual recordings due to privacy reasons but all obtained data.

Here you can find the answers to the questionnaire.

Here you can find the features we retrieved from the audio recordings like the SFF, loudness, duration etc.