Back to search

Aristotle University Of Thessaloniki (AUTH)

Development of acoustic signal processing applications on embedded systems

Abstract

dc:description

The recording and processing of acoustic signals has contributed significantly to technological progress across a broad range of scientific and industrial fields, from medical diagnosis and structural integrity assessment to industrial fault detection, ecoacoustics and bioacoustics, and multimedia information retrieval. Recent advances in artificial intelligence, including speech recognition, generation and autonomous navigation, have further extended the capacity to exploit acoustic information. The present thesis develops three practical applications in these domains, spanning experimental design, measurement acquisition, algorithm design, embedded implementation and evaluation under real-world conditions, with emphasis on algorithms suited to low-cost, low-power hardware and on comparing conventional statistical methods with machine learning models. The first application concerns leak detection in industrial metallic pipelines, applicable to unseen installations with a strict zero false alarm requirement. From accelerometer recordings of noise and artificial leaks, a threshold-definition method was developed from a single reference measurement, without training on leak data. An algorithm based on the root-mean-square value of the signal was designed first, with adaptive thresholds for the varying noise level. Binary and one-class classification models were then examined, the latter attaining the highest accuracy at the cost of retraining needs. The final method extends the root-mean-square algorithm to three features, reducing false alarms and improving detection speed, implemented on a wireless microprocessor node. The second application concerns honeybee colony health monitoring, aimed at identifying the queenless state. A measurement system recorded two colonies continuously over two years, healthy and artificially queenless. Spectral analysis revealed a marked acoustic change from the first hour of the queen's absence. A conventional algorithm estimates colony state as a degree of certainty from distribution overlap, revealing the principal spectral bands of the queenless state and the exact moment of swarming. Low-complexity neural networks combined with principle component analysis were trained in parallel. Both methods were implemented on an Arduino Zero, trading processing time for memory in the statistical method against faster, more memory-intensive neural-network inference. The third application concerns Byzantine hymn identification through audio fingerprinting. As hymns are vocal, performed live by different chanters at different tempi and scales with no single reference recording, fingerprints must be robust to noise, tempo and scale variation. On a dataset of 351 executions of nine hymns, a simple spectral-peak method was compared with one based on the two-dimensional Fourier transform of the constant-Q spectrogram, yielding fingerprints invariant to tempo and scale, matched via correlation matrices and evaluated with and without artificial noise. The final client-server implementation was optimised and its performance characteristics were assessed on a Raspberry Pi 4, showing practical applicability on low-capability hardware. Overall, the three applications confirm that demanding acoustic processing techniques, designed for computational efficiency and noise robustness, can be implemented on embedded systems and operate reliably in real-world conditions, with practical application to industrial, environmental and cultural problems.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Kampelopoulos, Dimitrios
  • Καμπελόπουλος, Δημήτριος

Subjects

dc:subject × 14

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/62941

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Kampelopoulos, Dimitrios; Καμπελόπουλος, Δημήτριος. Development of acoustic signal processing applications on embedded systems. Aristotle University Of Thessaloniki (AUTH), 2026. http://hdl.handle.net/10442/hedi/62941