AI based Music Recommendation System Using Deep Learning

Project Title

AI based Music Recommendation System Using Deep Learning

Project Area of Specialization Artificial IntelligenceProject Summary

The customized recommendation framework for music should accurately represent private tastes. To obtain tailored feedback for the needs of various viewers, it takes adjustments. To find the

better  deep  learning  model  for  the  recommendation  may  pave  a  way  for  a  better  recommender.

Compared to the previous era, with commercial music streaming sites that can be downloaded from mobile devices, digital music availability is currently plentiful. It takes a very long time to figure out all this digital music and induces data exhaustion. It may be helpful to create a music recommendation system that  can automatically scan  the music libraries  and suggest  appropriate songs to  users. The music  provider  will  anticipate and  then  give their  customers  the appropriate  songs  based  on  the

characteristics of the music previously heard by using the music recommendation system. Our Project would like to build a framework for music recommendations that can provide recommendations based on  the  similarity  of  audio  signal  features.  This  project  uses  the  Convolutional  Neural  Network

(CNN)  and the  Recurrent Neural  Network  (RNN).  Customized recommendation  system  for music should effectively represent private preferences. To attain tailored recommendations for the demands of different listeners, it needs changes and therefore, attempting to find a better deep learning model for the recommendation will pave a way for a better recommender.

Project Objectives

1. To find the better  deep  learning  model  for  the  recommendation of music.
2. To create a music recommendation system that  can automatically scan  the music libraries  and suggest  appropriate songs to  users.
3. To provide a system that can provide recommendations based on  the  similarity  of  audio  signal  features.

Project Implementation Method

CNN was inspired by the visual cortex of the brain and prominently used for object or face detection and classification. CNN have  three-dimensional layers height,  width and  depth and  all neurons  are not  connected to  the previous  layer rather a  layer is  only connected  to a  small portion  of neurons  in the previous layer.  LSTM is  also a  deep learning algorithm to selectively remember patterns for long duration of time and to overcame long term dependencies problem of RNN.  LSTM just does  some changes  to information by  implementing multiplication and  addition.In this work,  a deep learning  prototype is recommended  for the automatic  classification of  different of audio  signals. The proposed CNN and LSTM model has an end-to-end architecture with MFCC feature extraction methods. As we all know that CNN understands the data accurately using an image. So, at first, we need to convert our audio data into image. To visualize the data, plot the amplitude versus the time graph but it becomes harder to study the response so the frequency graph utilized for that purpose using fast Fourier transform  and we notice that data becomes more reluctant at low frequency so further we would down-sample the data (say to maybe 16Khz).

CNN was inspired by the visual cortex of the brain and prominently used for object or face detection and classification.

CNN have  three-dimensional layers height,  width and  depth and  all neurons  are not  connected to  the previous  layer

rather a  layer is  only connected  to a  small portion  of neurons  in the previous layer.  LSTM is  also a  deep learning

algorithm to selectively remember patterns for long duration of time and to overcame long term dependencies problem

of RNN.  LSTM just does  some changes  to information by  implementing multiplication and  addition.In this work,  a

deep learning  prototype is recommended  for the automatic  classification of  different of audio  signals. The proposed

CNN and LSTM model has an end-to-end architecture with MFCC feature extraction methods.

As we all know that CNN understands the data accurately using an image. So, at first, we need to convert our audio data

into image. To visualize the data, plot the amplitude versus the time graph but it becomes harder to study the response

so the frequency graph utilized for that purpose using fast Fourier transform  and we notice that data becomes more

reluctant at low frequency so further we would down-sample the data (say to maybe 16Khz).

Benefits of the Project

The  recommender system obtains using both CNN and RNN overall has better performance.

User don’t have to worry about finding the music he like on the internet, he gets the music he want through this system.

we conclude that our CNN  and RNN definition can operate till the  prototype is trained

carefully  to categorize  the  audio  signals.

Technical Details of Final Deliverable

The music  provider  will  anticipate and  then  give their  customers  the appropriate  songs  based  on  the characteristics of the music previously heard by using the music recommendation system. This  project  uses  the  Convolutional  Neural  Network (CNN)  and the  Recurrent Neural  Network  (RNN).  Customized recommendation  system  for music

should effectively represent private preferences.

Final Deliverable of the Project Software SystemCore Industry MediaOther Industries IT Core Technology Artificial Intelligence(AI)Other Technologies OthersSustainable Development Goals Good Health and Well-Being for PeopleRequired Resources

Elapsed time in (days or weeks or month or quarter) since start of the project	Milestone	Deliverable
Month 1	start	problem
Month 2	agile	meetings
Month 3	coding of program	June 2022
Month 4	checking of the project	Checking without errors
Month 5	presentation and thesis	final round examination