بررسي عوامل موثر در درك گفتار و ارائه يك روش تركيبي براي افزايش ميزان درك سيگنال گفتار نويزي

Studying on Speech Intelligibility and Presenting a Hybrid Method for Intelligibility Improvement of Noisy Speech

In this paper, the problem of speech intelligibility has been addressed. Effective parameters on speech intelligibility are categorized as: noise, reverberation, bandwidth, and phonemic structure. Due to the dominant role of noise in speech intelligibility, this research concentrates on the noise and introduces a hybrid method for improving the intelligibility of noisy speech. Firstly, for each frame of noisy input, a voiced/unvoiced decision is made by peak picking on the autocorrelation sequence. Based on this decision, appropriate noise shaping filter is applied on the input signal. As the next process, the formant frequencies are estimated and then, the frequency band around each formant is amplified. In this part, a sequential procedure has been employed to segment the frame spectrum into bands containing one formant. The performance of the proposed method has been evaluated through objective and subjective tests. These tests totally prove the capability of the proposed system in intelligibility improvement of noisy speech.