Orca call analysis for acoustic separation: final 2 months

My Gsoc experience

Second and third months of coding out of 3

Summary of the project

I’m working on pre-processing the audio data in this project before submitting them to Spleeter, where Devdoot isolates the orca voices from the backdrop. I’ve been experimenting with various de-noising techniques in order to remove as much of the underwater noise as possible (eg. ships motors, wave crashing, other marine mammals sounds). I concentrated on several Fourier Transform algorithms throughout the first month. In the second month, I employed Discrete Wavelet Transforms, and in the last month, I concentrated on spectrogram de-noising.

During these months, we had the opportunity to meet a variety of experts in the field that work with Orcasound. We first met with Tom Denton, who presented his avian studies at Google. Then we met the HALLO group, where each GSOC student talked about their work informally. Finally, each GSOC contributor gave a presentation during the workshop “Towards Predicting Whale Movement: A Workshop in Harnessing the Power of Underwater Hydrophone Data for Detecting Whales in an Era of Forecast Models, Machine Learning, Cluster-based Computing, and Citizen Science Engagement”, hosted by the Department of Fisheries and Oceans (Canada), Simon Fraser University, Dalhousie University and Carleton University.

Meeting with Tom Denton and the HALLO Team

It was thrilling to be able to speak with them.

Meeting with Tom

During the meeting with Tom, I mentioned briefly that I was researching variances in tempo (as in songs) to distinguish boat motors out. He recommended that I read an article in which they identified boreal chorus frogs based on their pulses. This suggestion astounded me because I was unaware of any similar animal study.
Tom also discussed his professional experience and departmental adjustments. I had no prior experience with animal research or working with animal data before joining Orcasound. I was so inspired after hearing Tom’s story that I persuaded my university supervisors that I wished to work on a project using animal data for my dissertation.

This is a screenshot of the meeting with Tom
Here is a screenshot where I showed Tom the software that Devdoot and I created

Meeting with the HALLO team

On the other hand, there were significantly more individuals at the HALLO team meeting. It was nice to learn more about the work of other students. The HALLO team offered very technical suggestions and questions, providing a totally distinct perspective.

This is a screenshot of the meeting with the Hallo team

Presentation at the HALLO workshop

We each prepared a 5-minute presentation explaining what we do during the workshop. Here is the link of my presentation. One of the most challenging parts of preparing the presentation for me was determining how much details to provide. I wanted to make my slides simple so that the audience wouldn’t be too focused on reading the screen, while also showing them what I was referring to. Finally, I opted to show fewer graphs and instead play the de-noised results directly.

Second, there were some unexpected technical challenges. Unfortunately, certain audio clips that were playing on my laptop were unable to make a sound.

Finally, the time zone difference was quite tough. My presentation took place at about midnight (Central European Timezone), whereas Vancouver was at 15.30. It was amusing to wear a smart shirt while having pajama pants!

Here is a screenshot where I presented my work


The toughest obstacle I had to overcome was realizing that what I initially thought to be a boat sound was actually a humpback. I spent a month working on a specific audio file where I was aware that humpbacks were present alongside with orcas but was unsure of the sound that humpbacks made. I showed my supervisor Val the track during our meeting, and to my amazement, he pointed out that the “boat” was a humpback. When I learned that, I realized my assumptions about the regularities in tempo in boat motor sounds were simply wrong. Fortunately, Val revealed how a boat truly makes noise, allowing me to resume my study of actual boats.


To sum up, I had a great time working on this project. I was able to work with a fantastic team and also learned a lot about marine mammals. It has opened the door to my newfound love of animal research.
I’m going to keep working on this project and am eager to start more.

This is the link to the repository I have been working on.

Leave a Reply

Your email address will not be published. Required fields are marked *