Inspiration
My friends and I were in the car when we noticed that every time we had a conversation, someone (usually in the backseat) would ask to lower the volume of the sound system so that he/she could be involved in the conversation as well. Later on, once the conversation had ended, there was a repetitive awkward moment where one of us had to once again increase the volume to restore normality.
What it does
Adaptive Aud.io revolutionizes the social music listening experience. The music's volume will automatically be adjusted based on whether or not it detects nearby conversation.
How we built it
We built it using a custom made Voice Activity Detection software to detect whether people were talking or not after cleaning up the audio source using a Normalized Least Means Squares. This had to be done because we needed a means of separating voice input from music loop back and ambient noise. We also utilized Fast Fourier transformations in an attempt to further normalize the audio input.
Challenges we ran into
1) There was very little information on the Internet about echo cancellation (the technique we used to separate the voice from the music), and little information pertaining to receiving the output of the speaker (without using the mic). 2) Multi-Threading was very tricky since we had to use a total of 4 threads to efficiently run our program (reading in data from two audio streams, one for analyzing the data, and one for managing all of these). 3) Compiling everything together, as we had worked solo on individual aspects of the project. This was surprisingly much harder than expected, as the data had to be transformed for certain processing.
Accomplishments that we're proud
This was a major project for 4 freshman to be undertaking, especially since this was only our second Hackathon. We are proud that we were actually able to tackle all of the issues despite the little assistance that was available on the Internet. None of us had experience with signal processing and especially no experience with processing with sound byte information. Through extensive research during the first few hours of the Hackathon, we were able to successfully code the difficult aspects of the program, such as splitting the audio into the conversation and music.
What we learned
Through 36 hours of sweat, tears, and not showering, we were able to learn elementary signal processing techniques and how to combine them to achieve our goal, which was detecting voice. We also learned a great deal about teamwork, as we were consistently supporting each other throughout our challenges.
What's next for Adaptive Aud.io
Now that we have a working proof of concept, we hope to expand the project into creating a real software product which could be used by speaker companies to make smarter speakers. We plan on going full-out on this idea, as we are all very optimistic about the project and all believe it could have a big impact on the social music listening experience. Major industries for potential inclusion of Adaptive Aud.io include simple headphones, cars, and the Internet of Things.
Log in or sign up for Devpost to join the conversation.