1. Web-Based Music Genre
Classification for Timeline Song
Visualization and Analysis
Under the guidance of
Mr MOHUMMAD ABDUL WAHED
MADASU UMA MAHESH 20W91A05C0
MARILLA DEVENDER 20W91A05B6
K MANOJ KUMAR 20W91A05A2
GATLA SNEHA SRI 20W91A0565
2. Table of Contents:
• Abstract
• Existing System and its limitations
• Proposed System
• Advantages and Disadvantages
• System Requirements
• Software Architecture
• Modules
• Conclusion
3. Abstract
This is an web application that retrieves songs from YouTube and classifies them
into music genres. The tool explained in this study is based on models trained
using the musical collection data from Audioset. For this purpose, we have used
classifiers from distinct Machine Learning paradigms: Probabilistic Graphical
Models (Naive Bayes), Feed-forward and Recurrent Neural Networks and Support
Vector Machines (SVMs). All these models were trained in a multi-label
classification scenario. Because genres may vary along a song's timeline, we
perform classification in chunks of ten seconds. This capability is enabled by
Audioset, which offers 10-second samples. The visualization output presents this
temporal information in real time, synced with the music video being played,
presenting classification results in stacked area charts, where scores for the top-10
labels obtained per chunk are shown.
4. Existing System and its Limitations
Binary Pattern, which is the best result on this dataset. In 2016 Costa et.al worked on Music genre
recognition considering different datasets and their spectrograms. He trained the classifiers with
various techniques like Gabor filters, Local Binary Patterns, Phase Quantization, to extract
spectrograms and attained state-of-the-art outcome on numerous audio database. He has done the
comparison of results produced by Convolutional Neural Network (CNN) with the outcomes
procured by handcrafted features. His research was done on three different audio datasets with
definite attributes, specifically a western music dataset (ISMIR 2004 Database), a Latin American
music (LMD dataset), and African music dataset. His research proved that the CNN is best
alternative as it is favorable to all classifiers for music genre recognition.
5. Limitations
Subjectivity and Ambiguity: Genre classification inherently involves subjective judgment, and
different listeners may interpret genres differently. Additionally, some songs may fall into multiple
genres, making accurate classification challenging.
Limited Transparency: This lack of transparency makes it difficult to understand how certain
decisions are made, and users may not have insight into the training data and algorithms.
Dependency on User Interaction: Songs with fewer interactions may not benefit as much from
collaborative filtering, potentially leading to less accurate genre predictions for niche or less popular
genres.
Cultural and Regional Bias: The classification system may reflect biases based on the predominant
genres in certain regions or demographics, potentially overlooking or misclassifying music from less
represented genres.
Privacy Concerns: It collects user data for personalized recommendations, which can raise privacy
concerns. Some users may be uncomfortable with the amount of data collected and how it is utilized.
6. Proposed System
The visualization output presents this temporal information in real time, synced with the music
video being played, presenting classification results in stacked area charts, where scores for the
top-10 labels obtained per chunk are shown. We briefly explain the theoretical and scientific
basis of the problem and the proposed classifiers. Subsequently, we show how the application
works in practice, using three distinct songs as cases of study, which are then analyzed and
compared with online categorizations to discuss models performance and music genre
classification challenges.
7. Advantages and Disadvantages
Advantages:
Accessibility
Large Databases
Real-time Updates
User Interaction
Integration with Music Streaming
Services
Disadvantages:
Subjectivity
Ambiguity and Hybrid Genres
Cultural and Regional Variations
Dependency on Metadata
Privacy Concerns
Algorithmic Bias
8. System Requirements
HARDWARE REQUIRMENTS :
System : i3 or above
Ram : 4GB Ram.
Hard disk : 40GB
SOFTWARE REQUIRMENTS:
Operating system : Windows
Coding Language : python
9. Software Architecture
Fig: Classification flow. Given a video ID and a segment, the audio is downloaded, pre-processed by
converting it to MFCCs, then to VGGish embeddings, and finally it is passed to the models for classification.
14. Data Collection : Gathers a diverse dataset of audio files for training and testing the genre
classification model.
Data Preprocessing : Prepares the collected data for training the classification model.
Machine Learning Model Training : Train a machine learning model for music genre
classification.
Model Evaluation : Assess the performance of the trained model.
Timeline Song Visualization : Presents the music genres in a timeline or chronological format.
User Feedback and Correction : Allow users to provide feedback on genre classifications and
correct any misclassifications.
Retraining Pipeline : Continuously improve the model by incorporating user feedback.
Integration with Music Database/APIs : Enhance the system by integrating with external
music databases or APIs.
Security and Privacy : Address potential security and privacy concerns related to user data.
Deployment and Scalability : Ensure the system can handle varying loads and is deployable.
Modules
15. Conclusion
This project is a web application to discover music genres present in a song, along its timeline, based
on a previous experimentation with different machine learning models. By identifying genres in each
10-second fragment, we can get an idea of how each model perceives each part of a song. Moreover,
by presenting those data in a stacked area timeline graph, the application is also able to quickly show
the behavior of the models, which at the same time, is an interesting way to detect undesired or rare
predictions. It is also a first step towards an eventual user-centered MGC tool, in which the users can
submit feedback about the correctness of the predictions. To our knowledge, there is no visual tool
that provides this level of verification on genre classification results for different fragments of the
song.