Authors: Kokkinis, Elias K.; Mourjopoulos, John
When recording multiple musicians with close microphones in a real space, each microphone captures both the sound of the intended instrument and the undesired interference from other sources. The blind source separation (BSS) framework can be employed in order to suppress the leakage components. However typical convolutive BSS algorithms—attempting to blindly identify and invert the acoustic mixing system, which comprises several room impulse responses—do not achieve the desired result. In comparison, a Wiener filter was found to achieve superior performance in the separation of two audio sources, while being simple and computationally efficient.
Authors: Waterschoot, Toon van; Moonen, Marc
Acoustic feedback in sound reinforcing systems is often solved using a notch-filter-based howling suppression technique that depends on having an accurate means of detecting feedback. Many detection algorithms that are based on a single criterion have an unacceptable high false–positive rate when used for music. By combining multiple detection criteria, a novel algorithm reduced the false alarm rate from as high as 33% to as low as 3%. Careful adjustment of various thresholds in the algorithm is required to maximize sound quality. This study provides a unifying framework for evaluating howling detection.
Authors: Gillett, Philip; Johnson, Marty; Perini, Efrain; Toso, Alessandro; Harris, Daniel
Hearing protection headsets can be made acoustically transparent to the user if environmental sound can be easily localized. A proposed array of microphones processed by a filter-and-sum technique can recreate the target head-related transfer functions. A three-dimensional numerical model of a KEMAR mannequin with headset was experimentally verified to a 12-kHz bandwidth. The optimal six-microphone array (three per headphone) was measured in an anechoic chamber, and preliminary subjective tests were encouraging.
Authors: Chung, Li-Jing; Sen, Deep; Epps, Julien
Band-limited audio can be enhanced by artificially generating the missing spectral information even when there is no knowledge of the missing spectrum. The proposed algorithm analyzes multiple bands of the existing audio for their mutual information in order to estimate the most salient features of the missing components: gain, spectral envelope, and harmonicity. Objective and subjective tests show that the proposed algorithm is superior to existing schemes but with its harmonicity estimation module requiring high-computational cost due to high-order linear prediction filters.
Authors: Rumsey, Francis
[Feature] Forensic audio analysis now involves the use and analysis of digital techniques and signals in order to support criminal investigations. Tasks such as the detection of edits or the use of data compression can be made easier using sophisticated statistical analysis of audio samples. Voices may be disguised either intentionally or otherwise, and audio signal processing can be used either to simplify the segmentation of recordings into different talkers or to evaluate the effects of voice processing. There is also the question of evaluating the quality or intelligibility of signals that have been subject to heavy degradation, where traditional quality prediction algorithms may not work satisfactorily.
Institutional Subscribers: If you would like to log into the E-Library using your institutional log in information, please click HERE.
