Hi Grigorii, Welcome to the Shogun community! (please always cc the mailing list and introduce yourself to the community)
Thank you for you interest and the project idea.Yes this could be a nice project but it's difficult to judge from the vague description. To make sure your application could be successful, please note the steps in our “How to get involved”: - contribute Code (it is important for us to see Code contributions from you *before* the application deadline. This helps us to judge both your dedication and also expertise so we know, e.g. who might be best to mentor you. WE DO NOT ACCEPT APPLICATIONS WITHOUT PRIOR CODE CONTRIBUTION) - don’t just contact individual mentors but get active in the community (answer questions, report bugs, contribute) (Sorry I haven't been in IRC a while) - write a well thought-through project proposal. While it is great that you have already an idea, and have already worked in it, what you need for a GSoC project is a *detailed* plan (including, importantly, a timeline) that you can follow through with. I am happy to give you feedback on your application before the deadline. Looking forward to your contributions and your project proposal! Best wishes, Lea On Sun, 25 Mar 2018, 00:55 Grigorii Guz, <[email protected]> wrote: > Hello Lea, > > Sorry that I'm writing you here, I tried contacting you via IRC but when I > query your name it says that this nick doesn't exist. > My name is Grigorii, I'm a 3rd-year computer science student from > Vancouver. I'm planning to apply to GSoC for Data applications project. The > goal of my project is to create a system which, given a sample of audio and > the number of speakers on this sample, transcribes this sample and tells > which user is speaking at a given point of time (diarization) > > I have already done some progress with speaker diarization while working > on this problem in my university's software engineering club.I already have > a pipeline for generating the proper dataset from audio samples and a > simple model that gives decent accuracy for diarization. It has 95% > accuracy when distinguishing between 5 speakers on a dataset where each > datapoint is a set of features for 1 second of audiofile. Then, when > switching to different dataset with different set of speakers (transfer > learning), we get the similar accuracy. I will provide sample code with my > application. Do you believe that this would be a good project for summer? > > Have a good day, > Grigorii >
