Speech Enhancement Using Wiener Filter Based on Voiced Speech Probability

In this digitized world, quality, accuracy and adaptability are more emphasized. Due to immense practical applications, desire for clean signal is highly essential at the user end. In this work, speech signal is considered for enhancement. For this, Wiener filter is proposed based on voiced speech probability (VSP). The probability of the speech signal depends on the performance of voice activity detection (VAD). The decision directed method with likelihood ratio test estimates the noise which improves the performance of VAD. After finding the speech probability, the noise is updated and estimated. The mean square error is optimized by Wiener filter, and the signal is enhanced. For verification and comparison, signal-to-noise ratio (SNR) and perceptual evaluation of speech quality (PESQ) are considered. This proposed method can be utilized in real-time applications.