Personal Accomplishments
I am able to split the song into chunks, and I run song detection on each of the chunks. Currently chunks are split by the 10000 frames. The documentation for Librosa is confusing, and I spent alot of time understanding what the library was helping accomplish. I am able to get MFCC coefficients for a few chunks of 5 second intervals, and relative beat differences for those 5 second intervals.
This allows us to win back more time as we dont need to extract features in every chunk iteration. We also want split signal processing into another thread as this does not depend on song selection and spotify audio features. We are able to process each individual chunk effectively, and notice beat differences. I used asyncio functions to create individual tasks, and have them run concurrently. Debugging concurrent tasks is challenging because you dont know where the problems arise, and chunking’s formatting had to match up with what Shazam API reads, and further librosa reads.
Understanding MFCC in Librosa was challenging because of frequency band related information it gets, and for the longest time debugging frames, second intervals, and window_sizes, and understanding their subtle variations was always confusing.
(Polling spotify features only once when two consecutive chunks match up)
(Beat differences)
On Track?
I am on track. I spent alot of time debugging errors with chunking, and signal processing. I hope to have a MVP working signal processing by the end of this week latest by Wednesday, so that we can pass these chunk coefficients to the lighting engine. It is tough to say how long integration will take but for now we can pass some dummy values even if full integration is not possible.
Goals for Next Week
MVP for signal processing with 6-7 prominent features tested, and passed into other engines. We will spend more time fleshing out the show class, and testing the subsystems independently and in conjunction, so that we do not have any unexpected behavior ahead of the interim demo. I also want to test it on a few different samples, and see if we can get exaggerated differences at least for some of these features.