Repozytorium publikacji - Politechnika Gdańska

A method of speech intelligibility improvement in public address (PA) systems installed in reverberant acoustic spaces is proposed. The method is mainly intended for systems operating in high background noise levels. The algorithm is based on the near end listening enhancement approach. Signal-to-noise ratio (SNR) is evaluated in the octave frequency bands. Signal levels in the acoustic space are estimated using the measured impulse response of the room. Noise level is measured with the microphone, also a method of noise level estimation in presence of speech is proposed. Based on the calculated SNR values, a set of gain limits is selected, which allows for optimal utilization of the available acoustic power. Finally, gain values for the octave bands are calculated and applied to the speech signal. The overall performance of the proposed method was evaluated in a series of experiments performed in real acoustic spaces. Speech intelligibility was estimated using the STIPA index, measured with and without an additional, high-level noise source. The results indicate that the proposed method improved speech intelligibility in all tested cases, compared with the default PA system, especially in presence of a high-level background noise. The proposed method may be integrated with PA systems equipped with a digital signal processor.

Autorzy

• dr hab. inż. Józef Kotus link otwiera się w nowej karcie ,
• dr hab. inż. Grzegorz Szwoch link otwiera się w nowej karcie