fzy
|
**$100 at stake** Implement client-...
Star this Commitment
fzy commits to:
My paper is made public on 19th Dec (probably US time) and the one thing that can make my app stand out from other is the use of machine learning. I commit to implement client-side automatic segmentation using a pre-trained Recurrent Neural Network. Penalty for missing this deadline is $100 PER DAY going to my referee
No more reports due
|
My Commitment Journal
 fzy
December 22, 2019, 4:56 AM
|
|
| As of today my paper still has not been published..... Looks like I could've given me some more time to polish it up... but it's ok | |
|
fzy
December 19, 2019, 10:10 AM
|
|
| Yesterday I was able to finish the goals proposed for the day before. Today I was able to make the model run on real data and perform segmentation. Tomorrow's the last day, very likely I will complete the task. However I'll have to cut down on bells and whistles since the progress was much slower than I anticipated. | |
|
fzy
December 17, 2019, 12:29 AM
|
|
| The goal for today is to A) client-side able to load the model and its parameters B) extract spect or log_spect (mfcc to be done later) by the client-side C) Make sure the extracted value is the same as the one extracted by the python code on the server-side | |
|
fzy
December 16, 2019, 11:25 PM
|
|
| So AudioContext in WebAudioAPI has a predefined sample rate which cannot be changed (48k in Chrome in most modern computer, but could change drastically from 16k - 192k depending on the browser) so resampling is necessary. I was exploring options to predefine the sampling rate before loading audio such that there's only one resample necessary. I spent yesterday to try different options and it turns out to be incredibly messy. So it's a day wasted. | |
This Commitment has no photos.
Displaying 1-1 of 1 result.
December 16 to December 20
Successful
Success
Success
timwnz
- Referee approval report
Awesome. Progress every day.
fzy
- Committed user success report
.
