what about the accuracy?

[David-Jin]

@giulbia Hi, have you test the accuracy? What's machine learning method do you use?

[giulbia]

Hi, I used a Support Vector Machine. The accuracy score is 96% on this dataset.

[prashanthkolaneru]

hi, can you provide model building using data. i want to add some recordings and run it

[lishuchao012]

@giulbia hi, I test on your baby dataset accuracy score is more than 96%, on urban is 90%, but on ESC-50 is only 40%, can you explain why low in ESC-50?

[giulbia]

Hi @lishuchao012,

Can you give me more details please? What exactly have you tried? How was your dataset built? Because in ESC-50 there are a lot of categories (like "siren" for example) that can be misclassified.

Thanks

[prashanthkolaneru]

Good evening sir i had build the model using data of baby cry,laugh,noise , silence. I preprocessed this data and extract the features(mfcc,ste,logfilterbank) I had given this features to build ANN and got accuracy 98%. Finally i import this model to raspberry pi. I got stuck when i am giving live data from microphone here i am getting wrong detection. Example when i play baby cry its predicting zero Some times in silience predicting 1 Please help me to get accurate predictions with live data Thanking you Prashanth

On Tue, 31 Jul 2018 at 6:53 PM Giulia notifications@github.com wrote:

Hi @lishuchao012 https://github.com/lishuchao012,

Can you give me more details please? What exactly have you tried? How was your dataset built? Because in ESC-50 there are a lot of categories (like "siren" for example) that can be misclassified.

Thanks

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/giulbia/baby_cry_detection/issues/4#issuecomment-409193436, or mute the thread https://github.com/notifications/unsubscribe-auth/AeMLXX2rkirtVFPMJOOje1uelCxlhx2gks5uMEUogaJpZM4NSveY .

[lishuchao012]

@giulbia Thanks for your reply, I test in ESC-50 categories, there are 50 categories. Maybe the samples are too few in every category. It is not your model fault.

[giulbia]

@lishuchao012, actually my first attempt was on ESC-10 and the accuracy was already low, so I'm not that surprised that it is not good on ESC-50. Back then I noticed that many sounds, like cat's meow or breaking glass were confused with baby cry, and that's why I reduced the classification problem to 4 categories. These 4 categories look coherent enough in the context of baby sleeping in a bedroom. The full dataset is probably to big in the sense of number of categories but too small in the sense of training examples.

@raspberian1234 have you checked the quality of your recordings? What microphone are you using?

[prashanthkolaneru]

sir now everything is working fine, but i am facing wrong predictions when i play other sounds like(MUSIC)

can you help me in this i have taken dataset of four catageries (baby laugh,silience,cry,noise)

On Mon, Aug 13, 2018 at 3:01 AM, Giulia notifications@github.com wrote:

@lishuchao012 https://github.com/lishuchao012, actually my first attempt was on ESC-10 and the accuracy was already low, so I'm not that surprised that it is not good on ESC-50. Back then I noticed that many sounds, like cat's meow or breaking glass were confused with baby cry, and that's why I reduced the classification problem to 4 categories. These 4 categories look coherent enough in the context of baby sleeping in a bedroom. The full dataset is probably to big in the sense of number of categories but too small in the sense of training examples.

@raspberian1234 https://github.com/raspberian1234 have you checked the quality of your recordings? What microphone are you using?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/giulbia/baby_cry_detection/issues/4#issuecomment-412367995, or mute the thread https://github.com/notifications/unsubscribe-auth/AeMLXXCmmZffY7pWE0pJ7fjAds3BHIE2ks5uQImwgaJpZM4NSveY .

[giulbia]

Music is the kind of signal that is not present at all in the training set so it's normal that it makes a mistake. But recognising a song is not the goal of this model so why would you try that? If you really want the model to recognise "music" you should build a specialised one with a pertinent training set. You also want to be more specific about what you mean by "music" as there are so many different kinds.

[prashanthkolaneru]

My main focous is to build a model to detect only baby cry in realtime. The model shouldn’t detect other than it.

On Wed, 15 Aug 2018 at 4:30 PM Giulia notifications@github.com wrote:

Music is the kind of signal that is not present at all in the training set so it's normal that it makes a mistake. But recognising a song is not the goal of this model so why would you try that? If you really want the model to recognise "music" you should build a specialised one with a pertinent training set. You also want to be more specific about what you mean by "music" as there are so many different kinds.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/giulbia/baby_cry_detection/issues/4#issuecomment-413144333, or mute the thread https://github.com/notifications/unsubscribe-auth/AeMLXR6DcGcKyQOSj-B01EXuEj52ZTjfks5uQ-oggaJpZM4NSveY .

[prashanthkolaneru]

Sir do you have a dateset of adult cry I want to classify between baby cry and adult cry. Thanking you, Prashanth

On Wed, 15 Aug 2018 at 4:35 PM Prashanth kolaneru < prashanthpro.pro@gmail.com> wrote:

My main focous is to build a model to detect only baby cry in realtime. The model shouldn’t detect other than it.

On Wed, 15 Aug 2018 at 4:30 PM Giulia notifications@github.com wrote:

Music is the kind of signal that is not present at all in the training set so it's normal that it makes a mistake. But recognising a song is not the goal of this model so why would you try that? If you really want the model to recognise "music" you should build a specialised one with a pertinent training set. You also want to be more specific about what you mean by "music" as there are so many different kinds.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/giulbia/baby_cry_detection/issues/4#issuecomment-413144333, or mute the thread https://github.com/notifications/unsubscribe-auth/AeMLXR6DcGcKyQOSj-B01EXuEj52ZTjfks5uQ-oggaJpZM4NSveY .

[giulbia]

No sorry I don't have it.

Regards

[cojangee]

@lishuchao012, actually my first attempt was on ESC-10 and the accuracy was already low, so I'm not that surprised that it is not good on ESC-50. Back then I noticed that many sounds, like cat's meow or breaking glass were confused with baby cry, and that's why I reduced the classification problem to 4 categories. These 4 categories look coherent enough in the context of baby sleeping in a bedroom. The full dataset is probably to big in the sense of number of categories but too small in the sense of training examples.

@raspberian1234 have you checked the quality of your recordings? What microphone are you using?

Good evening i had build the model using data of baby cry,laugh,noise , silence from data file

Finally i import this model to raspberry pi. I got stuck when i am recoding live data from microphone here i am getting wrong detection. Example when i play baby cry its predicting zero

/// this is my microphone // https://www.amazon.com/Microphone-Computer-Laptop-Desktop-Driver/dp/B078J9BTMF

Please help me to get accurate predictions with live data Thanking you

[giulbia]

Hi @devilsun1

Have you checked the quality of the recordings? I had the same microphone and the quality was really poor. I had much better results (and positive predictions) using this one which is much bigger and more expensive than the little one I bought.