Nature and source of the csv of music

[mducoffe]

Hello,

is it possible to know the nature of the music you used (raw sound, MFCC, Fourier transform), the duration and their source ?

Thanks ! =)

[RobRomijnders]

Hi mducoffe,

thanks for reaching out.

The music is just raw .mp3 files. For stereo channels, I only used one of
the channels.
With a downsample factor of 30, the sample frequency is 1470Hz . No further
pre-processing.

If you like, I can send you the files. That way you can reproduce the
findings.

What are your thoughts on this? I;m working with time-series classification
all summer. I;m happy to hear about your project.

Kind regards,

Rob

On 21 June 2016 at 15:31, mducoffe [email protected] wrote:

Hello,

is it possible to know the nature of the music you used (raw sound, MFCC,
Fourier transform), the duration and their source ?

Thanks ! =)

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1, or mute the thread
https://github.com/notifications/unsubscribe/APbLxTpzSkBqP0A_yGlw90zcIuUMQO2zks5qN-ekgaJpZM4I6vGF
.

[mducoffe]

Hey,

thank you for your fast reply !

I am no expert about time serie, my phd is mostly about active learning and
multimedia classification for deep learning.
But one of my intern is working about CNN deconvolution on sound rather
than on image. There is a recent paper about deconv net for sound at ISMIR
2015 but it is on STFT
and I would like to check how 'good' may be the quality of the deconv
reconstruction.
But for that I need a CNN on sound, and theirs is not as good as yours.
plus you have trained on raw data which may get my job much easier =)

I would be happy if I can have the files, but thank you already for the
details, l will definitely use your architecture.

Cheers

Mélanie

2016-06-21 17:35 GMT+02:00 RobRomijnders [email protected]:

Hi mducoffe,

thanks for reaching out.

The music is just raw .mp3 files. For stereo channels, I only used one of
the channels.
With a downsample factor of 30, the sample frequency is 1470Hz . No further
pre-processing.

If you like, I can send you the files. That way you can reproduce the
findings.

What are your thoughts on this? I;m working with time-series classification
all summer. I;m happy to hear about your project.

Kind regards,

Rob

On 21 June 2016 at 15:31, mducoffe [email protected] wrote:

Hello,

is it possible to know the nature of the music you used (raw sound, MFCC,
Fourier transform), the duration and their source ?

Thanks ! =)

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1, or mute the
thread
<
https://github.com/notifications/unsubscribe/APbLxTpzSkBqP0A_yGlw90zcIuUMQO2zks5qN-ekgaJpZM4I6vGF

.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#1 (comment),
or mute the thread
https://github.com/notifications/unsubscribe/AKb5QO0BbMwYZTMB8SjbRfePFDA_YfKrks5qOATVgaJpZM4I6vGF
.

[RobRomijnders]

Hey Melanie,

Wow, multimedia classification for deep learning. That sounds like an
interesting PhD.

Your question made me reopen the project. I made some improvement to the
batch normalizer. Now it gets to 87% accuracy. The new code is on Github.

The new commit also contains the data_music.csv. Now you can rerun the
entire script yourself.

Let me know what you think. I;m happy to help

Moreover, you must know I am looking for an internship next winter. If you
have something interesting at your lab, I'm eager to talk.

Rob

On 21 June 2016 at 18:05, mducoffe [email protected] wrote:

Hey,

thank you for your fast reply !

I am no expert about time serie, my phd is mostly about active learning and
multimedia classification for deep learning.
But one of my intern is working about CNN deconvolution on sound rather
than on image. There is a recent paper about deconv net for sound at ISMIR
2015 but it is on STFT
and I would like to check how 'good' may be the quality of the deconv
reconstruction.
But for that I need a CNN on sound, and theirs is not as good as yours.
plus you have trained on raw data which may get my job much easier =)

I would be happy if I can have the files, but thank you already for the
details, l will definitely use your architecture.

Cheers

Mélanie

2016-06-21 17:35 GMT+02:00 RobRomijnders [email protected]:

Hi mducoffe,

thanks for reaching out.

The music is just raw .mp3 files. For stereo channels, I only used one of
the channels.
With a downsample factor of 30, the sample frequency is 1470Hz . No
further
pre-processing.

If you like, I can send you the files. That way you can reproduce the
findings.

What are your thoughts on this? I;m working with time-series
classification
all summer. I;m happy to hear about your project.

Kind regards,

Rob

On 21 June 2016 at 15:31, mducoffe [email protected] wrote:

Hello,

is it possible to know the nature of the music you used (raw sound,
MFCC,
Fourier transform), the duration and their source ?

Thanks ! =)

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1, or mute the
thread
<

https://github.com/notifications/unsubscribe/APbLxTpzSkBqP0A_yGlw90zcIuUMQO2zks5qN-ekgaJpZM4I6vGF

.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<
#1 (comment)
,
or mute the thread
<
https://github.com/notifications/unsubscribe/AKb5QO0BbMwYZTMB8SjbRfePFDA_YfKrks5qOATVgaJpZM4I6vGF

.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#1 (comment),
or mute the thread
https://github.com/notifications/unsubscribe/APbLxW88IzETONB2X3fC76-B3yafhwBkks5qOAvLgaJpZM4I6vGF
.

[mducoffe]

Good evening Rob, thanks for your help
I told my supervisor Frederic Precioso about you and he says he will have
fundings for an internship. If you want to discuss this with it, you can
send your cv to both our professional adress :
[email protected]
[email protected]

I will definitely look into your code next week, thanks again for sharing =)

Cheers

Mélanie Ducoffe

2016-06-21 19:38 GMT+02:00 RobRomijnders [email protected]:

Hey Melanie,

Wow, multimedia classification for deep learning. That sounds like an
interesting PhD.

Your question made me reopen the project. I made some improvement to the
batch normalizer. Now it gets to 87% accuracy. The new code is on Github.

The new commit also contains the data_music.csv. Now you can rerun the
entire script yourself.

Let me know what you think. I;m happy to help

Moreover, you must know I am looking for an internship next winter. If you
have something interesting at your lab, I'm eager to talk.

Rob

On 21 June 2016 at 18:05, mducoffe [email protected] wrote:

Hey,

thank you for your fast reply !

I am no expert about time serie, my phd is mostly about active learning
and
multimedia classification for deep learning.
But one of my intern is working about CNN deconvolution on sound rather
than on image. There is a recent paper about deconv net for sound at
ISMIR
2015 but it is on STFT
and I would like to check how 'good' may be the quality of the deconv
reconstruction.
But for that I need a CNN on sound, and theirs is not as good as yours.
plus you have trained on raw data which may get my job much easier =)

I would be happy if I can have the files, but thank you already for the
details, l will definitely use your architecture.

Cheers

Mélanie

2016-06-21 17:35 GMT+02:00 RobRomijnders [email protected]:

Hi mducoffe,

thanks for reaching out.

The music is just raw .mp3 files. For stereo channels, I only used one
of
the channels.
With a downsample factor of 30, the sample frequency is 1470Hz . No
further
pre-processing.

If you like, I can send you the files. That way you can reproduce the
findings.

What are your thoughts on this? I;m working with time-series
classification
all summer. I;m happy to hear about your project.

Kind regards,

Rob

On 21 June 2016 at 15:31, mducoffe [email protected] wrote:

Hello,

is it possible to know the nature of the music you used (raw sound,
MFCC,
Fourier transform), the duration and their source ?

Thanks ! =)

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1, or mute the
thread
<

https://github.com/notifications/unsubscribe/APbLxTpzSkBqP0A_yGlw90zcIuUMQO2zks5qN-ekgaJpZM4I6vGF

.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<

#1 (comment)

,
or mute the thread
<

https://github.com/notifications/unsubscribe/AKb5QO0BbMwYZTMB8SjbRfePFDA_YfKrks5qOATVgaJpZM4I6vGF

.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<
#1 (comment)
,
or mute the thread
<
https://github.com/notifications/unsubscribe/APbLxW88IzETONB2X3fC76-B3yafhwBkks5qOAvLgaJpZM4I6vGF

.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#1 (comment),
or mute the thread
https://github.com/notifications/unsubscribe/AKb5QKozpqdIkvOwRvk_wfGS0rFEbgFbks5qOCF8gaJpZM4I6vGF
.

[RobRomijnders]

Dear Melanie,

Thank you for offering this opportunity.
This week I also received two offers to work with deep learning on medical
data.
As I have a medical background, these projects suit me better.

I hope you find another qualified intern for your project.

Concerning the code, I am happy to explain more or provide help if you
decide to use it.

With kind regards,
Rob

On 23 June 2016 at 21:36, mducoffe [email protected] wrote:

Good evening Rob, thanks for your help
I told my supervisor Frederic Precioso about you and he says he will have
fundings for an internship. If you want to discuss this with it, you can
send your cv to both our professional adress :
[email protected]
[email protected]

I will definitely look into your code next week, thanks again for sharing
=)

Cheers

Mélanie Ducoffe

2016-06-21 19:38 GMT+02:00 RobRomijnders [email protected]:

Hey Melanie,

Wow, multimedia classification for deep learning. That sounds like an
interesting PhD.

Your question made me reopen the project. I made some improvement to the
batch normalizer. Now it gets to 87% accuracy. The new code is on Github.

The new commit also contains the data_music.csv. Now you can rerun the
entire script yourself.

Let me know what you think. I;m happy to help

Moreover, you must know I am looking for an internship next winter. If
you
have something interesting at your lab, I'm eager to talk.

Rob

On 21 June 2016 at 18:05, mducoffe [email protected] wrote:

Hey,

thank you for your fast reply !

I am no expert about time serie, my phd is mostly about active learning
and
multimedia classification for deep learning.
But one of my intern is working about CNN deconvolution on sound rather
than on image. There is a recent paper about deconv net for sound at
ISMIR
2015 but it is on STFT
and I would like to check how 'good' may be the quality of the deconv
reconstruction.
But for that I need a CNN on sound, and theirs is not as good as yours.
plus you have trained on raw data which may get my job much easier =)

I would be happy if I can have the files, but thank you already for the
details, l will definitely use your architecture.

Cheers

Mélanie

2016-06-21 17:35 GMT+02:00 RobRomijnders [email protected]:

Hi mducoffe,

thanks for reaching out.

The music is just raw .mp3 files. For stereo channels, I only used
one
of
the channels.
With a downsample factor of 30, the sample frequency is 1470Hz . No
further
pre-processing.

If you like, I can send you the files. That way you can reproduce the
findings.

What are your thoughts on this? I;m working with time-series
classification
all summer. I;m happy to hear about your project.

Kind regards,

Rob

On 21 June 2016 at 15:31, mducoffe [email protected] wrote:

Hello,

is it possible to know the nature of the music you used (raw sound,
MFCC,
Fourier transform), the duration and their source ?

Thanks ! =)

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1, or mute the
thread
<

https://github.com/notifications/unsubscribe/APbLxTpzSkBqP0A_yGlw90zcIuUMQO2zks5qN-ekgaJpZM4I6vGF

.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<

#1 (comment)

,
or mute the thread
<

https://github.com/notifications/unsubscribe/AKb5QO0BbMwYZTMB8SjbRfePFDA_YfKrks5qOATVgaJpZM4I6vGF

.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<

#1 (comment)

,
or mute the thread
<

https://github.com/notifications/unsubscribe/APbLxW88IzETONB2X3fC76-B3yafhwBkks5qOAvLgaJpZM4I6vGF

.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<
#1 (comment)
,
or mute the thread
<
https://github.com/notifications/unsubscribe/AKb5QKozpqdIkvOwRvk_wfGS0rFEbgFbks5qOCF8gaJpZM4I6vGF

.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#1 (comment),
or mute the thread
https://github.com/notifications/unsubscribe/APbLxZoDSSo6sPdrnbLUIxC72fNcexOqks5qOuArgaJpZM4I6vGF
.