We are so excited to finally open up Genderfluent to the public! We got a lot of great feedback from our initial beta testers. While our initial private beta test took longer than expected, it became clear that we needed to make some fundamental improvements to the app. What did we need to work on?
Algorithm improvements
Improving the gender and pitch estimation algorithms was what took most of the time. We had to go back to the drawing board and build completely new algorithms, since the ones we released during the private beta were underperforming. One new feature is providing confidence values for the gender estimations (i.e. if the app estimates that your voice sounds female, how confident is it in that estimation?). We found that when the recorded audio volume and proportion of audio containing speech were too low, the gender estimates were getting unacceptably inaccurate. Lastly, we made some additional performance improvements to the pitch estimation process to improve accuracy and decrease power/battery usage. That last part is particularly helpful for our smartphone users.
Visualization improvements
We also greatly improved the visualization of pitch and gender estimates, as well as adding an additional visualization for audio volume. These new visualizations are faster, more interactive, and can now let you zoom and pan.
You’ll notice that Genderfluent is still in a beta version. It’s our way of communicating to our users that the app is still new and may still have some fundamental issues. However, we will continually make improvements and we are eager to get any feedback from our users, since it is one of the main drivers for making the app better for everyone.
That’s it for now. In the meantime, try it out and let us know how you like it!