DSP Tools
Welcome to this section of our website, where we will delve into the fascinating world of Digital Signal Processing (DSP) tools and their applications in our work. We aim to provide you with a comprehensive understanding of how each DSP tool plays a pivotal role in our projects, allowing us to tackle complex challenges and achieve remarkable results.
Frequency Domain Analysis & Computation using
Fourier Transform (in)
Our project used Frequency domain in two ways:
- One way was to use the frequency domain to view audio data in a different domain to glean insights.
- The other usage was for applying filters and signal transformations in the frequency domain. In some cases, such as the notch filters used to model the pinna-spectral effect, it was easier to work in the frequency domain so we did a frequency-domain multiplication rather than a time-domain convolution.
Convolution & LTI Systems of Impulse Response (in)
In our database method, after generating the HRTF from the database, we convolve the left and right channels with their respective HRTFs to obtain the output.
Head-Related Transfer Function
The Head-Related Transfer Function (HRTF) is the crux of our project and described with better treatment at this page.
Filtering/Filter Design
We incorporated filter design into our model for the pinna-related spectral notches. We made 4 notches filters (a type of filter that cuts off a thin notch around a desired frequency) at several high frequencies that had their magnitudes adjusted by elevation-dependent functions.
Correlation (for Interaural Time Difference) (out)
Correlation is a statistical-related method often used in signal processing to measure how similar two signals are, which can be used for measuring efficacy of a routine, recognizing patterns in signals, or in our case, compare time delays in the left and right channels. As the data is not guaranteed to have full equivalent left and right channels (which is actually very unlikely due to empirical fluctuations), we cannot simply assume that both channels will be exactly the same, and thus we use correlation to figure out the sections that are *nearly* the same.