Title

AWS re:Invent 2022 - Build and deploy a live, ML-powered music genre classifier (BOA322)

Summary

The session focused on building and deploying a live machine learning (ML) powered music genre classifier.
The speakers were Sohan Maheshwar, Linda, and Suman Devnath, all from AWS, with backgrounds in development advocacy and data engineering.
The use case presented was for DJs to maintain the vibe of an event by sequencing songs of similar genres.
The GTZAN dataset was used for training the ML model, which contains labeled audio files across 10 different genres.
Tools used included scikit-learn, Librosa for audio file analysis, and TensorFlow for image representations of audio files.
Features such as spectrograms, spectral centroids, and chroma were extracted from audio files using Librosa.
The ML model was built and trained using Amazon SageMaker, which offers various benefits like built-in algorithms, AutoML, and scalability without upfront hardware costs.
The trained model was saved and uploaded to an S3 bucket.
The deployment was serverless, using AWS Lambda functions, Amazon API Gateway, and Amazon EFS.
AWS SAM (Serverless Application Model) was used to automate the deployment process.
The session concluded with a demonstration of the entire process from dataset to deployment and encouraged attendees to explore the GTZAN dataset and SageMaker workshops.

The GTZAN dataset's consideration for audio quality and bias reduction is crucial for building an unbiased ML model.
The use of Librosa for feature extraction demonstrates the importance of domain-specific libraries in ML workflows.
Amazon SageMaker's features, such as built-in algorithms and AutoML, simplify the ML model building process, making it accessible to developers without deep ML expertise.
The serverless deployment approach emphasizes the trend towards minimizing operational overhead for developers, allowing them to focus on building and scaling applications without managing infrastructure.
AWS SAM's role in simplifying and automating the deployment of serverless applications highlights the importance of Infrastructure as Code (IaC) in modern cloud environments.
The session showcased the end-to-end process of building an ML application on AWS, from data preparation to model training and deployment, providing a practical example of how AWS services can be integrated to solve real-world problems.