maigoakisame / cmu-thesis Goto Github PK

Hi Maigo,

How do you get filterbank features from audios? I didn't find any code related to data processing and the data downloaded from the bash file is already prepared.

Hi, thank you for your great work!
when I run the ./download.sh
I got the following:
Connecting to islpc21.is.cs.cmu.edu(islpc21.is.cs.cmu.edu)|128.2.190.16|:80... failed: Connection time out.

I checked the computer is connected to the Internet.
Is that something wrong with the your sever?

@MaigoAkisame Thanks for your work. I have 2 questions below, hope you can help me figure out.

1. input 400 frames, the params: win_length and hop length (maybe 64ms per frame with 25ms hop_length)
2. the input size fixed to 400 frames, if the input audio lasts 1s, can the net work? or if I have a audio lasts 10s, I want to predict the label every 1s, does it work?

Hi, Maigo
I'm trying to utilize a DCASE winning system on Audio Tagging Challenge (https://github.com/multitel-ai/urban-sound-tagging), which uses your work on TALNet as pretrained model.
But I'm getting forbidden messages from CMU server, it seems all links that begin with "islpc21.is.cs.cmu.edu" are not working, is there any other way to get the model?

The link you provide seems cannot use, can you share a new download path? If the audioset data is too large, you can just share the DCASE2017 data. Thanks.

https://github.com/lijuncheng16/AudioTaggingDoneRight/blob/b8ec7f509b2a4909777e7572445a0f6233392ce1/src/utilities/stats.py

The AUC calculation would return a different result VS. using the sklearn metrics package, very interesting, but haven't figured out which step was different.

Great work!
How can I get your pytorch version of the SoundNet code and the pretrained model of your phd thesis?