Wiener Filter Based Noise Reduction Algorithm with Perceptual Post Filtering for Hearing Aids

• Published in: International journal of image, graphics and signal processing Vol. 11; no. 7; pp. 69 - 81
• Main Authors: Pujar, Rajani S, Kulkarni, Pandurangarao N
• Format: Journal Article
• Language: English
• Published: Hong Kong Modern Education and Computer Science Press 01.07.2019
• Subjects: Algorithms; Amplitudes; Background noise; Compression tests; Consonants (speech); Dynamic range; Filter banks; Frequencies; Hearing aids; Hearing loss; Intelligibility; Listening; Masking; Noise; Noise reduction; Sentences; Signal to noise ratio; Spectra; Speech; Speech recognition; Splitting; Wiener filtering; Words (language)
• ISSN: 2074-9074; 2074-9082; 2074-9082
• DOI: 10.5815/ijigsp.2019.07.06

This paper presents a filter bank summation method to perform spectral splitting of input signal for binaural dichotic presentation along with dynamic range compression coupled with noise reduction algorithm based on wiener filter. This helps to compensate the effect of spectral masking, reduced dynamic range, and improves speech perception for moderate sensorineural hearing loss in the adverse listening conditions. We have considered cascaded structure of noise reduction technique; Filter Bank Summation (FBS) based amplitude compression and spectral splitting. Wiener filter produces the enhanced signal by removing unwanted noise. The signal is split into eighteen frequency bands, ranging from 0-5KHz, based on auditory critical bandwidths. To reduce the dynamic range, amplitude compression is carried out using constant compression factor in each of the bands. Subjective and objective assessment based on Mean Opinion Score (MOS) and Perceptual Evaluation of Speech Quality (PESQ) scores, respectively, are used to test the Perceived quality of speech for different Signal-to-Noise Ratio (SNR) conditions. Vowel Consonant Vowel (VCV) syllable /aba/ and sentences were used as the test material. The results of the listening tests showed MOS scores for processed speech sentence “sky that morning was clear and bright blue” (4.41, 4.2, 3.96, 3.6, 3.08 and 2.66) as compared with unprocessed speech MOS scores ( 4.53, 1.21, 1.16, 1.06, 0.8, 0.483) for SNR values of ∞, +6, +3, 0, -3 and -6 dB respectively, and PESQ values (Left Channel: 2.6192, 2.5355, 2.5646, 2.5513, 2.5221, and 2.4309; Right Channel: 2.5889, 2.3001, 2.3714, 2.4710, 2.3636, and 2.4712) for SNR values of ∞, +6, +3, 0, -3 and -6 dB respectively, indicating the improvement in the perceived quality for different SNR conditions. To evaluate the intelligibility of the perceived speech, listening test was carried out for hearing impaired (moderate Sensorineural Hearing Loss (SNHL)) persons in the presence of background noise using Modified Rhyme Test (MRT).The test material consists 50 sets of monosyllabic words of consonant-vowel-consonant (CVC) form with six words in each set. Each subject responded for a total of 1800 presentations (300 words x 6 different SNR conditions). Results of the listening tests (using MRT) showed maximum improvement of (27.299%, 23.95%, 24.503%, 23.602%, and 23.498%) in the speech recognition scores at SNR values of (-6dB, -3dB, 0dB, +3dB, +6dB) compared to unprocessed speech recognition scores. Reductions in response times compared to unprocessed speech response times at lower SNR values were observed. The decrease in response times at the SNR values of -6, -3, 0, +3 and+6 dB were 1.581, 1.41, 1.329, 1.279, and 1.01s, respectively, indicating improvement in intelligibility of the speech at lower SNR values.