Essay On Sound Event Recognition

4727 Words19 Pages
CHAPTER 1 INTRODUCTION The sounds that we hear in everyday environments contain a wide variety of acoustic information which assists us in the accomplishment of many daily tasks. Among these sounds, speech is arguably the most important as we use it extensively for communication. However, beyond speech, we are capable of producing non-speech sounds like crying, coughing or clapping which convey our emotion. In addition, there are more general sounds produced by objects, for example door closing or a phone ringing that provides us with more information and context about the environment. The project topic on sound event recognition (SER) aims to automatically detect and classify non-speech sounds to provide more information about the surroundings.…show more content…
Wold, T. Blum, D. Keislar, and J. Wheaten [4] illustrated sound event recognition by describing sound through a common set of characteristics, such as pitch, loudness, duration and timbre. Although common ASR features can capture these to an extent, they often do so implicitly, rather than designing a feature to capture a specific characteristic directly. For example, MFCCs capture loudness through the zeroth coefficient, and pitch and timbre are represented in the remaining coefficients. Hence, they may not provide the best representation for the full range of sound characteristics, despite providing a good SER baseline performance. This leaves scope to develop novel ways of representing the sound event information. In particular, timbre is important as it represents a range of distinctive characteristics about the sound. They capture elements such as the spectral brightness, roll-off, bandwidth and harmonicity. Brightness is defined as the centroid of power spectrum, while spectral roll-off measures the frequency below which a certain percentage of the power resides. Both are a measure of the high frequency content of the signal. Bandwidth captures the spread of the spectral information around the centroid, while harmonicity is a measure of the deviation of the sound event from the perfect harmonic
Open Document