Research Paper Review

Subject: Noise reduction

Topic: Assessing the Effect of Noise-reduction to the Intelligibility of Low-pass Filtered Speech

The residual hearing in hearing-impaired (HI) listeners is due to high frequency elimination in the words or sentences reaching the ear of these listeners. The residual hearing is only due to the low frequency components in the speech. This selective or residual hearing affects the speech intelligibility of the HI listeners. The hearing of these people degrades further in a noisy environment, because the low frequency signals are also affected due to the noise components in the surroundings. One of the solutions found out for enhancing residual hearing is through noise reduction in the signals reaching the ear of the HI people. This helps the low frequency components in high frequency speech can be detected to give highly maximum hearing capability to them.

The technique used was that low pass filtered noise- corrupted signals were used at cut-off frequencies 750 and 1000Hz. Low pass filtering was done using a linear phase FIR filter of order = 10 and sampling frequency of 16kHz. These LPF signals were processed by four single-channel noise reduction algorithms. The experiment was performed in a sound proof room. The subjects were asked to participate in a total of 20 conditions [ = 2 low pass cut off frequencies (1000 and 750Hz) x 2 SNR levels (-3dB and -6dB) x 5 signal processing conditions]. The intelligibility scored was defined as the ratio between number of correctly recognised words and total number of words combined in each list of 10 MHINT (Mandarin Hearing in Noise Test) sentences.

This work investigated the effect of four types of existing most-used single-channel noise reduction algorithms to the intelligibility of low-pass filtered Mandarin sentences. Low pass filtering simulated the effect of understanding speech with low frequency residual hearing of hearing-impaired listeners. Results showed that existing single-channel NR algorithms did not improve the intelligibility of low-pass filtered speech. Among the four types of single-channel noise reduction algorithms examined in this study, Wiener filtering had the least negative influence to the intelligibility of low-pass filtered speech.