Experiment Overview

Testing each participant

Participants were asked to sing the following series of 5 vowels: /ɑ/, /e/, /i/, /o/, and /u/. They were first given a starting note, which was A3. I also had the participants sing on a C4 and E4. However, many of the guys couldn’t hit one of both of these notes and ended up singing the octave down. Since I ended up needing to focus on just one note, I chose A3, since I had the most consistent data for that note.

Participants next heard a vocal recording of the same vowels at the same pitch sung by me. For each of the pitches, they heard the recording once, and then I played it back a second time for them to sing along to and try to blend with my vowels. 

I also repeated this experiment with the first two lines of “Amazing Grace” with the intention to consider blend in the context of an actual song, but this part similarly did not make the final cut. 

All of the previous recordings were recorded using an external microphone hooked up to my computer and recorded on tracks in Garageband. Participants wore headphones so that they could hear their own voice through the microphone as well as my voice for the blending part of the trials, while the microphone would only pick up their voice.

Analysis

The analysis of this data is further explained in the following pages on this website. However, I will give a brief overview of the process here.

Praat

I inputted the .wav files from Garageband into Praat and created TextGrid files in Praat in order to annotate the sounds. By slightly editing code provided by Joey Stanley on his website, I was able to add in a way to extract pitch. Additionally, instead of calculating formants based on midpoint, I did so using mean by editing the code further.

Python

I wrote Python code to pull data from the .csv files created by the Praat Script into Python code. I wrote functions to separate the data into solo and blend recordings, and I found the average difference between participant data and target data for all vowels and all formants. I then plotted the results in bar graphs using Matplotlib libraries.

Leave a Reply

Your email address will not be published. Required fields are marked *