| Description |
In practical environments, the quality and intelligibility of speech signals are often degraded by background noises generated from car-engine, traffic, aircraft cockpit, factory floor and multitalker babble in the cafeteria. Thus, the performance of speech processing related applications, such as, hearing aids devices, and speech recognition or speaker authentication systems is deteriorated drastically. In these systems, speech enhancement can be applied for improving the quality and reliability of services. The objective of this project is to develop a single channel speech enhancement system in practical environments based on a Wiener filter in conjunction with the masking properties of a psychoacoustic human auditory model. A human listener tolerates additive noise as long as it remains below a noise masking threshold. Based on this threshold, a double filter will be designed which will accentuate the denoising when noise is perceptually significant as well as involve activating Wiener filter even when noise is inaudible. A perceptual post processing method will also be developed for reducing the musical residual noise yielding from the Wiener denoising technique. As a consequence, the best tradeoff will be achieved among the amount of noise reduction, the speech distortion and the level of annoying musical noise in a perceptual sense. Students will first learn briefly about the basic terminologies in speech processing related to speech enhancement. For simulation they will use standard noise and speech databases and also their own recorded speech. Concentration will be given on the most important step, the noise masking threshold calculation, accuracy of which is susceptible to noise leading to a severe degradation in the enhancement performance. Students will not only implement the speech enhancement system but also perform objective and subjective evaluation of the proposed system in the presence of different environmental noises using temporal, spectral and perceptual criteria. |