Repository for training age detection models using the Common Voice dataset.
Active members of the team working on this repo include:
- Jim Schwoebel (Boston, MA)
- Rajesh Singh (Nashville, TN - Vanderbilt University)
Before downloading, make sure you have roughly 6 GB of hard disk space on your computer.
You can download the data by going to this link and clicking download (an arrow on the top right corner of page).
There are 3 folders: male, female, and other. The data take the form of .mp3 files. This data format can be directly read by third party libraries like librosa.
Here are some goals to beat with the data:
| Age model name | Accuracy | Standard Deviation | Modeltype |
|---|---|---|---|
| teens.pickle | 0.76209217575633 | +/- 0.0212424639835302 | knn |
| twenties.pickle | 0.747827586437699 | +/- 0.00483503800126851 | knn |
| thirties.pickle | 0.765463935070702 | +/- 0.00626231620665817 | knn |
| fourties.pickle | 0.774859823805254 | 0.0186105837039435 | knn |
| fifties.pickle | 0.794545131389912 | 0.0151656184818545 | knn |
| sixties.pickle | 0.797694462400345 | 0.0346864094031604 | knn |
| seventies.pickle | 0.795753660637382 | 0.0378425083662649 | hard voting |