Semester/Master Thesis: Audio Compression Algorithms for Wireless Vocal Monitoring

Introduction
Multiple research fields such as animal welfare research or studies of vocal learning require the monitoring of animal vocal behavior using animal-borne sensors, which are wireless because the use of cables to transfer the recorded data may affect the normal behavior of the animal. Using wireless technologies, large volumes of acoustic data can be collected with minimal impact on the animals’ mobility.
Our lab is currently working on several projects that requires the vocal monitoring of songbirds. Our current efforts are based on TinyBird (1), a small wearable sensor node for the continuous acoustic monitoring of birds. TinyBird communicates with the central receiver over Bluetooth Low Energy (2) (BLE) and is powered by coin batteries. As a good dataset requires long-lasting recordings, the power consumption must be minimized; communication is usually the most power-hungry subsystem in this type of devices.
One approach to minimize power is to reduce the amount of data transmitted over BLE without affecting the goodness of the received signal. The most common approach is data compression. In signal processing, data compression algorithms encode the information using fewer bits than the original representation. These methods are divided into two main categories: lossless and lossy compressions2. The former eliminates redundant information, the latter reduce bitrate by removing unnecessary or less important information2. Goal of the project
The objective of this project is to considerably reduce power consumption of TinyBird during streaming of audio data. To this aim, the student will explore several audio compression methods, both lossless and lossy. Afterwards, the student will implement on the SoC of TinyBird the ones that better meet our needs in terms of i) precision in pitch estimation, ii) accuracy in identifying vocalizations and iii) high compression ratio and percentage of energy saving.
Tasks of the project
1. Familiarization of lossless and lossy data compression methods
2. Implementation of a restricted group of selected algorithms on the target platform (SoC in use: nRF528324, Nordic Semiconductor)
3. Performance evaluation of the chosen algorithms with a significative focus on the power consumption
4. Reconstruction of the compressed signals in Python/MatlabPrerequisites
• Proficiency in C programming
• Good knowledge of Python or Matlab
• Knowledge about embedded systems
• Basic knowledge about Bluetooth
• Ability to work independently
• Motivation
Supervisors
Prof. Richard Hahnloser, rich (at) ini.ethz.ch, Institute of Neuroinformatics
Matilde Dirodi, mdirodi (at) ini.ethz.ch, Institute of Neuroinformatics Bibliography
[1] Oliver Brunecker and Michele Magno. “TinyBird: An Energy Neutral Acoustic Bluetooth-Low-Energy Sensor Node with RF Energy Harvesting”. In: ().doi:10.1145/3362053.3363498
[2] Learn about Bluetooth Low Energy: https://www.bluetooth.com/learn-about-bluetooth/tech-overview/
[3] L. A. Fitriya and T. W. Purboyo, "A Review of Data Compression Techniques," International Journal of Applied Engineering Research, vol. 12, no. 19, pp. 8956-8963, 2017
[4] nRF52832: https://www.nordicsemi.com/products/nrf52832