Using the 'Singer's Formant' to Reduce Inaccuracies in the Location of a Singer On Stage

摘要

There are many possible benefits to knowing the location of a singer onstage, for example for automatic tracking of the singer with lights or cameras. This paper examines the question of locating a singer by finding the time delay between the singing arriving at the radio and stage microphones. Traditional time delay estimation techniques, such as the generalised cross-correlation and phase correlation methods, produce a lot of incorrect results when used in this situation. These errors arise due to: (i) the nature and level of the background noise; (ii) the large distances between the microphones in question; (iii) the nature of the singing itself. This paper examines a means of reducing errors that arise when the signals received by the two microphones are not sufficiently similar to produce a correct result using correlation-based delay estimation techniques. In order to do this, the delay estimate is found only at sections of the signal containing a singer's formant, which is a significant peak in the region of 2.5 to 3kHz that occurs when singers project their voices above the masking frequencies of the accompanying music. While the number of incorrect results is reduced, the results show that the overall accuracy is not improved. This can be attributed to the fact that frames containing a singer's formant are narrowband signals, with high frequency spectral components, and are therefore highly susceptible to phase-wrapping errors.