Audio Classification on AudioMNIST dataset

[MahmoudMohajer]

Zama Bounty Program: Proposition

• Library targeted: Concrete ML

• Overview: Creating an audio classifier on AudioMNIST dataset using the Concrete ML library.

• Description:
  Sound data is a crucial yet sensitive information source that demands privacy and protection. Leaked audio can contain vital and confidential data. I propose a project aimed at integrating sound recognition and classification into the Concrete ML library's use-cases. This initiative intends to enhance Concrete ML's capabilities, making it adept at handling sound-related tasks securely. The goal is to build an audio classifier using the AudioMNIST dataset, thereby advancing the library's utility and contributing to the broader field of machine learning in privacy-preserving environments.

  Planned Milestones for One Month:

  • Week 1:

    • Familiarize with the AudioMNIST dataset and its nuances.
    • Research existing audio classification techniques and frameworks.
    • Set up the development environment with Concrete ML and necessary dependencies.

  • Week 2:

    • Preprocess the AudioMNIST dataset for model training.
    • Design a suitable neural network architecture for audio classification.
    • Begin the implementation of the audio classifier using Concrete ML.

  • Week 3:

    • Continue refining the audio classifier model.
    • Conduct rigorous testing and validation to ensure accuracy and reliability.
    • Address any issues or challenges encountered during the implementation process.

  • Week 4:

    • Finalize the implementation of the audio classifier on the Concrete ML platform.
    • Optimize the model for performance and efficiency.
    • Document the entire process, including codebase, architecture, and usage guidelines.
    • Prepare a comprehensive report detailing the project, its challenges, and solutions implemented.

• Reward: The proposed reward value for this bounty will be communicated after further discussion and validation by the Zama team. We believe that the complexity and significance of this project merit a competitive reward, commensurate with the efforts involved.

• Related links and references:
  • AudioMNIST Dataset
  • Research Paper
  • AudioMNIST using PyTorch

We are excited about the prospect of enhancing Concrete ML's capabilities and contributing to the realm of privacy-preserving machine learning. Looking forward to collaborating on this innovative and impactful project.

Best regards,
Mahmoud Mohajer

[MahmoudMohajer]

the proposal canceled because of similarity to the classic MNIST dataset.