Exploring Data through Sound - Victor Minces

Structured data vs. Unstructured data

  • Sound waves, signals

Sounds are made of frequencies

Signal Generator
Spectrogram
Pink Trombone - Vocal Cavity Simulator

Frequencies of the Human Voice

Fundamental Frequency around 80/100hz
Many, many harmonics (~100)
Filtered harmonics

Data Science and Sound - Akshay Nagarajan, Victor Minces

Sound Representations of Data

Two directions
Data Sound
Sound Data

Data Sonification

Geiger Counter (1908)

The Geiger counter, one of the earliest examples of data sonification, converts radiation levels into audible clicks. It contains a small amount of gas that reacts to radiation, generating ions. Each ion produces a tiny amount of electricity, which is then converted into sound. More clicks indicate higher radioactivity.

Cosmic Sonification

Bullet cluster, dark matter
Data showing dark matter are represented by the lowest frequencies, while X-rays are represented by the highest frequencies

Machine Listening

Machine Listening

Machine listening is a fascinating field that enables machines to interpret and understand audio signals, beyond speech recognition including environmental sounds, animal calls, etc.

  • Speech recognition: involves machines identifying and processing human speech to convert spoken words into text or commands
  • Music Information Retrieval: analyzing music to extract information
  • Environmental sound classification

Bestiary of Rhythms

Machine Listening Process

  • Data Collection
  • Preprocessing
  • Model training: often CNN or RNN
  • Testing and refinement

Tools for Machine Listening

  • TensorFlow and PyTorch
  • LibROSA

Data Collection in the Wild

Unstructured Bioacoustics

ARBIMON: classification based on distinct sound signatures for machine listening

Songlines as Memory Code

Naturalistic way of encoding memory in song lyrics