Ai-Yukino/smol-puhe

REAMDE.md for smol-puhe

• Extract estimates for the first four formants F1, F2, F3, and F4, and the fundamental frequency F0 from audio clips listed in validated.tsv from the Common Voice Corpus 9.0 - Finnish dataset using Praat
• Build a model which predicts the perceived gender of a speaker based on a relationship between the formants F1, F2, F3, F4, and the fundamental frequency F0
• Redo of puhe repo without audio files and with .gitignore which ignores any additional .mp3 and .wav files you add to your locally cloned repo

---

💬 Voice sample clips 💬

If you don't want to downlaod the full data set above, you can listen to and download the 10 sample clips we used in our analysis below:

Female speaker sample clips

• Female speaker 1 sample clip
• Female speaker 2 sample clip
• Female speaker 5 sample clip
• Female speaker 7 sample clip
• Female speaker 9 sample clip

Male speaker sample clips

• Male speaker 17 sample clip
• Male speaker 18 sample clip
• Male speaker 23 sample clip
• Male speaker 26 sample clip
• Male speaker 29 sample clip

🔗 Other links 🔗

• 📝 Docs for Common Voice Dataset

🔗🔗 Good introduction to formants

• 🎥 Speech Acoustics 4 - Source-filter model | Listen Lab (9 minutes)
  • Good starting place to build intuition about what formants are
• 🎥 Speech Acoustics 5 - vowel formants | Listen Lab (10 minutes)
  • More details about formants
• 🎥 Praat 3 - Formant settings | Listen Lab (7 minutes)
  • Motivation for why we need to manually tweak Praat parameters for each individual speaker

📦 Versions 📦

Ai's computer

• tidyverse 1.3.1
• R version 4.2.0 (2022-04-22 ucrt)
• RStudio 2022.02.3+492 "Prairie Trillium" Release (1db809b8323ba0a87c148d16eb84efe39a8e7785, 2022-05-20) for Windows Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) QtWebEngine/5.12.8 Chrome/69.0.3497.128 Safari/537.36
• Praat version 6.2.14 (May 24, 2022)
  • FastTrack plugin
    • Repo cloned on June 8, 2022
    • Last commit on that day is probably this