Abstract

In this work, five audio signal datasets: Speech On.wav (SO), Speech Off.wav (Sof), Speech Disambiguation.wav (SD), Speech Misrecognition.wav (SM) and Window Navigation Start.wav (WNS), were acquired from Microsoft Windows Sound Notification Collection to investigate the effects of signal amplitude on two frequency estimation methods, namely, Discrete Fourier Transform (DFT) and DFT resolution approaches. The signals were analyzed using MATLAB version R2018b. The DFT and DFT resolution estimates of the fundamental frequency of each sample was determined. The Power spectrum analysis of the audio signals was carried out to extract data about the signals. The amplitude of each of the signals was recorded. The DFT estimate values for the five audio signal datasets were 703.4180Hz, 349.2124Hz, 1176Hz, 347.0099Hz and 988.434Hz respectively, and their DFT Resolution estimates were 700.8324Hz, 349.1250Hz, 1177Hz, 343Hz, and 998.0526Hz while the amplitudes were 0.1236m, 0.1635m, 0.0222m, 0.1399m, and 0.0264m respectively. The DFT Resolution method outperformed DFT method with SO, Sof, SM by 2.5856Hz, 0.0874HZ and 4.0099Hz, respectively while DFT outperformed DFT Resolution with SD and WNS by 1Hz and 9.6186Hz respectively. The relationship between the amplitude of the signals and the performance of the estimation methods was examined by plotting a graph of amplitude against the improvement in DFT estimation methods. The application of DFT Resolution to the signals demonstrates a constant, albeit more pronounced, inverse relationship between amplitude and the estimated frequency. However, DFT methods performed better when analyzing signals with very low amplitudes as seen in the analysis of SD and WNS