Cleaning up sound

It may come as no surprise that when training a speech recognition system, it’s probably better to have high-quality audio. A high-quality clean audio file should have few acoustic artifacts which makes it easier to deal with than an audio file recorded in a noisy environment with a poor quality microphone. For example, the VoxForge corpus contains high-quality clear sound recordings, but the CHILDES corpus contains many old audio recording that lacks clarity. Also, because these recordings are of young children and their parents, these audio recordings are often inconsistent and surprising (not the good kind either).

So this week I worked on trying to remove unnecessary silences and denoise some of these artifacts so that we can have both proper training and testing data, and so that our acoustic analysis of the audio corpus is more robust.

There are a TON of ways to “clean” and equalize/compress a signal. For our purposes, we use MATLAB tools to help reduce the noisiness (variance) of these CHILDES recordings. Though, I do want to point out that many digital audio workstations (DAW) like Adobe Audition and Pro Tools offer premier noise reduction and sound analysis technologies. Again, I probably won’t use these, but I like Adobe Audition’s suite of tools:

More than interest

A couple of weeks ago I wanted to understand and describe the relationship between interest and learning. I believe this is an essential concept to the project. So, I tried to explain how a child’s interest is crucial for their language development. However, while interest is a vital component of any learning, I do think other motivations or mechanisms guide us to master all sorts of complex systems and concepts. For instance, the fear of appearing “dumb” (embarrassment & shame) and competition are also strong motivations for learning. Additionally, I find it inefficient and even inhibiting to only learn about one’s interest, as so much of our interest comes from implicit preferences. Even more, interest only develops one so much, there must be other mechanisms imposed by the environment that not only instruct but force one to grow.

Tapia Submission

This week Rebekah and I worked on finishing up our submission to the ACM Richard Tapia Celebration of Diversity in Computing Conference. Cool to see how our two projects are coming together.

Best,
EO