Deep learning for music & audio

A record of my work in the deep learning (DL) for audio and music course of my master's. Technically, deep learning is a part of ML, but ML basics were covered in a separate previous course.

Use of DL in audio & music

• Uncover features in audio representations (ex. image recognition on spectograms of audio files)
• Classify audio or parts of audio
• Musical analysis

---

To explore further, check the following:

• DL focused part
• Audio & music focused part

Programming support topics

• pytorch: PyTorch library for handling vectors & matrices as tensors
  • pytorch.tensor: Tensor
    • Tensor ranks
      • Rank 0: Floating point number
      • Rank 1: Array or vector
      • Rank 2: Matrix or vector of vectors
      • Rank 3: Matrix of matrices
    • Matrix multiplication with tensors (how is it done?)
    • Broadcasting
    • Memory management
      • Assignment only assigns new names & does not create copies
        • Hence, using assignment, different identifiers can be made to refer to the same memory
        • Assignment does not by itself duplicate data
      • In-place assignments can be done in the following ways (given tensors X and Y):
        • Y += X
        • Y[:] = Y + X
        • NOTE: Y = Y + X is not in-place & creates a new memory location for Y
    • Converting tensors to other Python datatypes (ex. NumPy array, float, etc.)
• Generator functions in Python
  • Keyword yield used to mark the point from where the function must continue in the next call
  • Local variables for the next call and execution resumes from yield