Skip to content
This repository has been archived by the owner on Jun 3, 2019. It is now read-only.

Vocal Pitch Research

Victor Carri edited this page Nov 19, 2018 · 31 revisions

Purpose

To determine how vocal pitch affects the recognition of lyrics, without anything else being altered.

Materials

  • Our Cleansio Python script
  • A Python script which takes 2 files as input and prints the number of differences between the files as output, created by me
  • Free Online Pitch Shifter
  • Genius Lyrics

Songs

  • The song "Please Release Me", by Englebert Humperdinck
  • The song "Ruby Tuesday", by Melanie
  • The song "Royals", by Lorde
  • The song "Starboy", by The Weeknd

Methods

  1. Fetch the correct lyrics of the song.
  2. Use the pitch shifter to shift the pitch of the song. I created 4 varieties:
  • Shifted up 1 semitone
  • Shifted up 2 semitones
  • Shifted down 1 semitone
  • Shifted down 2 semitones
  • And the original song
  1. Run Cleansio on a variation, producing Cleansio's guess at the lyrics.
  2. Use the similarity script which I created to print the # of differences between the official lyrics and the lyrics as output by Cleansio.
  3. Repeat steps 1-4 for each variation of the song.
  4. Repeat step 5 for each song.

Results

Song Artist Original voice type # of differences - original song # of differences - down 2 semitones # of differences - down 1 semitone # of differences - up 1 semitone # of differences - up 2 semitones # of words in the original song
Please Release Me Englebert Humperdinck Male baritone 45 (37.5%) 45 (37.5%) 45 (37.5%) 42 (35%) 48 (40%) 120
Ruby Tuesday Melanie Safka Female, contralto? 87 (47.2826086957%) 77 (41.847826087%) 101 (54.8913043478%) 66 (35.8695652174%) 80 (43.4782608696%) 184
Royals Lorde Female, soprano? 72 (19.2%) 78 (20.8%) 66 (17.6%) 78 (20.8%) 85 (22.6666666667%) 375
Starboy The Weeknd Male, contralto? 140 (31.25%) 172 (38.3928571429%) 59 (13.1696428571%) 59 (13.1696428571%) 59 (13.1696428571%) 448

Discussion

Shifting the song up 1 semitone reduced the number of differences in the first 2 cases and the last case, but shifting it down 1 semitone worked better in the case of the mostly-vocal "Royals".

Future Work

More tests must be done to determine whether the shifting the entire song 1 semitone up consistently reduces the number of differences, and how shifting the song up or down by more than 2 semitones will affect the number of differences.