In Class DSP Tools:

For our project we are using low-pass FIR filters, FFT analysis, and spectrograms. We are using the spectrograms to understand the types of sounds we have in our audio samples. We are using FFT analysis to understand how our signal looks in the frequency domain, and see the differences between male and female voices. We are using low-pass filters in parts of our speech processing.

 

Out of Class DSP Tools:

We are using two main tools for speech processing. The first is the cepstral transform. The cepstral transform is used to separate the voice and words characteristics of an audio signal. Specifically we are using this on short vocal sounds because they are easier to analyze with cepstral techniques. The other technique we are using is linear predictive coding. We are using this as another way of separating our audio signal into voice and words. We are then attempting to switch these signals between two audio signals in order to hear the words in different voices.

 

Coolest Thing:

While we were studying different speech processing techniques, we stumbled across a program called Praat. This program has tons of cool and useful tools for speech processing. It’s been fun to play around with this program, analyze formants and pitch, and ultimately hear one person's speech in another person's voice. We are trying to model our project around this concept, but it’s been hard to do effectively. This program also allows us to visualize speech in many different ways. Here is a sample plot from the program.