The procedure

[IcyHao]

Hi, I was confused about the procedure of the experiment although I had looked through the README.md. Could you list out the steps of the experiment? Thanks a lot.

[qiuqiangkong]

Hi, the easiest way is to download the pretrained models and try to inference an audio clip. This should be in the first part of the README.

[IcyHao]

Thanks for the clarification. Inside the runme.sh file, at the third line you define the workspace path. May I know what exactly it refers to? And also in the 43rd line of the same file, in the path of the audios_dir there is a file called "eval_segments". Are these refer to the audio clips?

[qiuqiangkong]

Hi, workspace is your local directory that will save checkpoints, etc. eval_segments refer to evaluation audio clips.

On Thu, 11 Feb 2021 at 22:49, IcyHao notifications@github.com wrote:

Thanks for the clarification. Inside the runme.sh file, at the third line you define the workspace path. May I know what exactly it refers to? And also in the 43rd line of the same file, in the path of the audios_dir there is a file called "eval_segments". Are these refer to the audio clips?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-777515876, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSIQKGFITW5KN6YMC73S6PU7LANCNFSM4WAUJBGQ .

[IcyHao]

When I try to download the evaluation wavs, I get the error saying "dataset.py download_wavs: error: the following arguments are required: --audios_dir". As I already include the audios_dir in my code when executing, I got no idea how I got this error. Do you face this problem before>

Below is my code -- python3 d:/audioset_tagging_cnn-master/utils/dataset.py download_wavs --csv_path="./datasets/audioset201906""/metadata/eval_segments.csv" --audios_dir="./datasets/audioset201906""/audios/eval_segments"

[qiuqiangkong]

Hi, I did not have this problem. Did you use windows? Check if the dir of of audios_dir is correct?

On Thu, 18 Feb 2021 at 07:48, IcyHao notifications@github.com wrote:

When I try to download the evaluation wavs, I get the error saying "dataset.py download_wavs: error: the following arguments are required: --audios_dir". As I already include the audios_dir in my code when executing, I got no idea how I got this error. Do you face this problem before>

Below is my code -- python3 d:/audioset_tagging_cnn-master/utils/dataset.py download_wavs --csv_path="./datasets/audioset201906""/metadata/eval_segments.csv" --audios_dir="./datasets/audioset201906""/audios/eval_segments"

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-780929532, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSJSIFZJX3NX2CMFEI3S7RIVRANCNFSM4WAUJBGQ .

[qiuqiangkong]

I saw there is an "" in your path "./datasets/audioset201906""/metadata/eval_segments.csv"?

On Thu, 18 Feb 2021 at 16:59, IcyHao notifications@github.com wrote:

I use windows. I had checked that it is in the correct directory. [image: Capture] https://user-images.githubusercontent.com/57998537/108332307-8d6b9280-71c7-11eb-816b-b81bdb42259a.JPG

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-781190987, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSMS36IQBNSCAEEULR3S7TJHDANCNFSM4WAUJBGQ .

[IcyHao]

Oh, it's my mistake. But then when I execute it, I got another error. Is this because the function cannot find the file?

OSError: [WinError 123] The filename, directory name, or volume label syntax is incorrect: '_logs/download_dataset/D:'

[qiuqiangkong]

'_logs/download_dataset/D:' is not a good path. There can be a few path issues on windows systems. I thought something like "D:/xxx/workspaces/" would be better.

On Thu, 18 Feb 2021 at 23:58, IcyHao notifications@github.com wrote:

Oh, it's my mistake. But then when I execute it, I got another error. Is this because the function cannot find the file?

OSError: [WinError 123] The filename, directory name, or volume label syntax is incorrect: '_logs/download_dataset/D:'

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-781445502, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSN3S4UHE3CVQPWQWTLS7U2IZANCNFSM4WAUJBGQ .

[IcyHao]

Inside the dataset.py, I found that there is a line that seems related to the error, is this what I need to change? But I had tried to change to '_logs/download_dataset/D:/audioset_tagging_cnn-master/{}' and it still gives me the same error.

logs_dir = '_logs/download_dataset/{}'.format(get_filename(csv_path))

[qiuqiangkong]

Try to print get_filename(csv_path) and see what you got?

On Tue, 23 Feb 2021 at 18:02, IcyHao notifications@github.com wrote:

Inside the dataset.py, I found that there is a line that seems related to the error, is this what I need to change? But I had tried to change to '_logs/download_dataset/D:/audioset_tagging_cnn-master/{}' and it still gives me the same error.

logs_dir = '_logs/download_dataset/{}'.format(get_filename(csv_path))

[image: 1] https://user-images.githubusercontent.com/57998537/108827686-d05da980-75bd-11eb-97b5-06b5fac196cd.JPG

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-784052502, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSN7QYDGI6F4IT3NVNDTAN4L3ANCNFSM4WAUJBGQ .

[IcyHao]

It shows D:\audioset_tagging_cnn-master\datasets\audioset201906\metadata\eval_segments

[IcyHao]

Ok, I got it. So after downloading the waves, I run the code of packing waveforms to the hdf5 file and it shows files do not exist. When I successfully download the evaluation wavs, I only got 10 files but the packing function is looking for 20000++ files. Is it because of I miss some steps or the download is not in the correct way?

0006.log

[qiuqiangkong]

Hi, check if the download is running correctly? Youtube is limiting the number of videos to be downloaded recently.

On Thu, 25 Feb 2021 at 07:31, IcyHao notifications@github.com wrote:

Ok, I got it. So after downloading the waves, I run the code of packing waveforms to the hdf5 file and it shows files do not exist. When I successfully download the evaluation wavs, I only got 10 files but the packing function is looking for 20000++ files. Is it because of I miss some steps or the download is not in the correct way?

0006.log https://github.com/qiuqiangkong/audioset_tagging_cnn/files/6039454/0006.log

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-785462814, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSJ4DX4LK5LZO4BM5MTTAWD4BANCNFSM4WAUJBGQ .

[IcyHao]

These are the only thing that I got when I download the evaluation wavs. Are these things that you suppose to get when running the download?

[qiuqiangkong]

The downloading looks not successful. There should be 20k+ files downloaded.

On Thu, 25 Feb 2021 at 17:54, IcyHao notifications@github.com wrote:

These are the only thing that I got when I download the evaluation wavs. Are these things that you suppose to get when running the download?

[image: download] https://user-images.githubusercontent.com/57998537/109135643-3463a780-774f-11eb-918e-ac897b42c5cc.JPG

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-785768595, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSIDWENIHK7E7KGTOWLTAYM35ANCNFSM4WAUJBGQ .

[IcyHao]

Just a quick question, what is the point of the evaluation set?

[qiuqiangkong]

The evaluation set contains around ~20k audio clips for evaluating the performance.

On Mon, 8 Mar 2021 at 07:11, IcyHao notifications@github.com wrote:

Just a quick question, what is the point of the evaluation set?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-792371704, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSJHLLZSBXKVSNKEYC3TCQBZ3ANCNFSM4WAUJBGQ .

[IcyHao]

So it is not used for training and inferencing?

[qiuqiangkong]

No, eval is not for training.

On Mon, 8 Mar 2021 at 16:44, IcyHao notifications@github.com wrote:

So it is not used for training?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-792583905, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSNCXTDWLEJQPGO2DZLTCSE7FANCNFSM4WAUJBGQ .

[IcyHao]

Sorry for interrupting you again. I had downloaded the balanced training segments and I found that they are not converted to wav forms. So I delete them and redownload again, but now it can't run the code at line 101 and below. To run the code, the length of video_paths needs to bigger than zero but what I print out is an empty set in line 99 (The picture below is the output when I run the download_wavs command. Do you know what is going on?

[qiuqiangkong]

Hi, have you checked if the videos are downloaded? If not, L101 will return empty.

On Sun, 14 Mar 2021 at 01:32, IcyHao @.***> wrote:

Sorry for interrupting you again. I had downloaded the balanced training segments and I found that they are not converted to wav forms. So I delete them and redownload again, but now it can't run the code at line 101 and below. To run the code, the length of video_paths needs to bigger than zero but what I print out is an empty set in line 99 (The picture below is the output when I run the download_wavs command. Do you know what is going on?

[image: Capture] https://user-images.githubusercontent.com/57998537/111038560-9dbb0a00-8421-11eb-8ee2-e66aed049fca.PNG

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-798672083, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSLWXQX5W5FXYXXFLDTTDOOSRANCNFSM4WAUJBGQ .

[IcyHao]

Hi again, the terminal had shown that it did not have enough RAM when I run the inference & training step. Other than reducing the training samples or the batch size, are there any ways that could reduce the usage of RAM?

[qiuqiangkong]

Hi, there are 3 ways to reduce RAM usage: 1) Reduce batch size, 2) Reduce model size, 3) Use a larger RAM.

Best wishes, Qiuqiang

On Thu, 25 Mar 2021 at 07:33, IcyHao @.***> wrote:

Hi again, the terminal had shown that it did not have enough RAM when I run the inference & training step. Other than reducing the training samples or the batch size, are there any ways that could reduce the usage of RAM?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/30#issuecomment-806250401, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSJR34RESR7H2IR43LLTFJZD7ANCNFSM4WAUJBGQ .

[IcyHao]

Ok, thanks a lot.