Masters Degrees (Electrical and Electronic Engineering)
Permanent URI for this collection
Browse
Browsing Masters Degrees (Electrical and Electronic Engineering) by Subject "Acoustic localization"
Now showing 1 - 3 of 3
Results Per Page
Sort Options
- ItemAcoustic detection of the short pulse call of Brydes whales using time domain features and hidden Marcov models(Stellenbosch : Stellenbosch University, 2020-03) Wache Ngateu, Gaelle Venessa; Versfeld, Daniel Jaco J.; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.ENGLISH ABSTRACT: The biological group of cetaceans is frequently studied nowadays as passive acoustic monitoring (PAM) is commonly used to extract the acoustic signals produced by cetaceans, in the midst of noise sounds made by either man during shipping, gas and oil explorations or by natural sounds like seismic surveys, wind and rain. In this research work, the acoustic signal of short pulse call of inshore Bryde's whales is detected using time domain features and hidden Markov models (HMM). HMM is deployed as a detection and classi cation technique due to its robustness and low time complexity during the detection phase. However, some parameters such as the choice of features to be extracted from the acoustic short pulse call of inshore Bryde's whales, the frame durations of each call and the number of states used in the model affect the performances of the automated HMM. Therefore, to measure performances like sensitivity, accuracy and false positive rate of the automated HMM; three time domain features (average power, mean and zero-crossing rate) were extracted from a dataset of 44hr26mins recordings obtained close to Gordon's bay in False bay, South Africa. Moreover, to extract these features the frame durations of each vocalisation was varied thrice; 1 ms, 5 ms and 10 ms. Also, the HMM used three different number of states (3 states, 5 states and 10 states) which were varied independently so as to evaluate the HMM. On an overall performance, the HMM yields best performances when it uses 10 states with a short frame duration of 1 ms and average power as the extracted feature. With regard to this, the automated HMM shows to be 99:56% sensitive, and dependable as it exhibits a low false positive rate of 0:1 with average power inferred as the best time domain feature used to detect the short pulse call of inshore Bryde's whales using the HMM technique.
- ItemKleinvoet: a spatially-distributed temporally-synchronised infrasonic recorder(Stellenbosch : Stellenbosch University, 2023-12) Geldenhuys, Christiaan; Niesler, Thomas; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.ENGLISH ABSTRACT: We present the hardware design of a temporally synchronised high-fidelity low-cost infrasonic-capable (6Hz) passive acoustic monitoring (PAM) device. This research instrument allows recordings made by independent spatially distributed nodes to be synchronised in time for the purpose of acoustic monitoring and localisation of the African elephant, using for example time difference of arrival (TDoA). Each recorder is capable of sampling with a 24-bit resolution at a sampling frequency that is adjustable between 8 kHz and 192 kHz. Audio samples are stored locally on a microSD card, along with global navigation satellite system (GNSS) derived timestamp metadata that is shown to be accurate to within 500 ns (6σ). Four prototype recorders were built and evaluated. Estimates of the acoustic frequency response were obtained using Welch spectral averaging and confirmed that the recorders have a flat (±3 dB) pass-band over the frequency range 30Hz to 450 Hz. An analysis of the sampling clock confirmed an oscillator bias within specification (±2.5 ppm) and an oscillator stability, estimated using Allan variance, that was substantially improved using a simple regular resynchronisation method based on the GNSS timestamp metadata. Finally, without additional post-processing, the temporal alignment error is to within half a sampling period. However, if additional signal processing techniques are used, this alignment error can be reduced to approach the GNSS timestamp accuracy of 500 ns, which is well below half the highest sample frequency. We conclude that the recorders are well suited for sparsely distributed PAM experimental data collection, especially over the frequency range of African elephant vocalisations, for which it is intended.
- ItemLearning to speak and hear through multi-agent communication over a continuous acoustic channel(Stellenbosch : Stellenbosch University, 2023-03) Eloff, Kevin; Kamper, Herman; Engelbrecht, Herman; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.ENGLISH ABSTRACT: Human infants acquire language in large part through continuous signalling with their caregivers. By interacting and communicating with their caregivers, infants can observe the consequences of their communicative attempts (e.g. through parental response) that may guide the process of language acquisition. We find many similarities between human language acquisition and the intuition of intrinsic motivation which serves as a basis of reinforcement learning. In contrast, current trends in natural language processing disregard this, instead focusing on having larger models and more data to learn the statistical relationships between words with none of the original goals of language in mind. Multi-agent reinforcement learning has proven effective for investigating emergent communication between social agents. Most of these studies, however, focus on communication with discrete symbols. Humans learn language over a continuous channel and language has evolved through gestures and spoken communication, both of which are inherently continuous. This channel is also time-varying: interactions take place in unique settings with different channel acoustics and types of noise. These intricacies are lost when agents communicate directly with purely discrete symbols. We therefore ask: are we able to observe emergent language between agents with a continuous communication channel? And if so, how does learned continuous communication differ from discrete communication? Our objective is to provide a platform to study emergent continuous signalling in order to see how it relates to human language acquisition and evolution. We propose a messaging environment where a Speaker agent needs to convey a set of attributes to a Listener over a noisy acoustic channel. This thesis makes two core contributions. Firstly, in contrast to recent studies on language emergence, we train our agents with deep Q-learning rather than REINFORCE. When using DQN, we show significant performance gains and improved compositionality. Secondly, we provide a platform to study spoken emergent language between agents. To showcase this, we compare discrete and acoustic emergent languages. We show that, unlike the discrete case, the acoustic Speaker learns redundancy to improve Listener coherency when longer sequences are allowed. We also find that the acoustic Speaker develops more compositional communication protocols which implicitly compensates for transmission errors over the noisy channel. In addition, we show early experiments with promising results in language grounding (to English) and effective generalisation to real-world communication channels.