The experimental work and analysis on implemented noise reduction algorithms for hearing aids are present here.
4.1 DWT and SS
Here, we analyzed the outcomes obtained after applying Discrete Wavelet Transform (DWT) and Spectral Subtraction (SS) Speech Enhancement techniques to enhance the quality of noisy audio signals for hearing aids. A discussion of the results obtained from the implementation of Speech Enhancement algorithms has been given below.
4.2 DWT-BSN
The efficiency of the proposed Discrete wavelet transform-balance sparsity norms (DWT-BSN) Speech Enhancement algorithm using a soft thresholding approach with multiple wavelet families, is demonstrated in Tables 6 and Figs. 9–10 respectively. Following post-denoising, the MSE of each and every Wavelet dramatically decreases, and the SNR increases, demonstrating a superb decrease in noise and enhancement of signal intelligibility. Wavelets like sym-15 and coif-4 exhibit the highest SNR improvements of 21.5293 dB and 21.3771 dB, respectively. From these experimental results, the DWT-BSN algorithm is a promising technique for Speech Improvement in Wavelet Transform-based hearing aids.
Table-6: Initial and denoised MSE and SNR across proposed Discrete wavelet transform - balance sparsity norms (DWT-BSN) wavelets
Wavelets
|
Ini-MSE
|
Den-MSE
|
Ini-SNR (dB)
|
Den-SNR (dB)
|
haar
|
0.0016590
|
0.0000952
|
5
|
17.1199
|
db_5
|
0.0016590
|
0.0000483
|
5
|
20.2840
|
db_10
|
0.0016590
|
0.0000469
|
5
|
20.4192
|
db_15
|
0.0016590
|
0.0000467
|
5
|
20.4460
|
sym5
|
0.0016590
|
0.0000487
|
5
|
20.2469
|
sym_10
|
0.0016590
|
0.0000473
|
5
|
20.3826
|
sym_15
|
0.0016590
|
0.0000458
|
5
|
21.5293
|
coif_3
|
0.0016590
|
0.0000482
|
5
|
20.2978
|
coif_4
|
0.0016590
|
0.0000474
|
5
|
21.3771
|
coif_5
|
0.0016590
|
0.0000470
|
5
|
20.4091
|
bior_5.5
|
0.0016590
|
0.0000553
|
5
|
19.6445
|
Table-7: Effects of Discrete wavelet transform - balance sparsity norms (DWT-BSN) approach on STOI and PESQ metrics accross Wavelets
Wavelets
|
Ini STOI
|
Den STOI
|
Ini PESQ
|
Den PESQ
|
haar
|
0.5167
|
0.9147
|
1.3067
|
4.1626
|
db-5
|
0.5167
|
0.9634
|
1.3067
|
3.9459
|
db-10
|
0.5167
|
0.9675
|
1.3067
|
3.6741
|
db-15
|
0.5167
|
0.9689
|
1.3067
|
3.5112
|
sym-5
|
0.5167
|
0.9637
|
1.3067
|
3.9501
|
sym-10
|
0.5167
|
0.9675
|
1.3067
|
3.6593
|
sym-15
|
0.5167
|
0.9687
|
1.3067
|
3.5447
|
coif-3
|
0.5167
|
0.9655
|
1.3067
|
3.8474
|
coif-4
|
0.5167
|
0.9672
|
1.3067
|
3.7297
|
coif-5
|
0.5167
|
0.9678
|
1.3067
|
3.6460
|
bior-5.5
|
0.5167
|
0.9643
|
1.3067
|
3.8664
|
The impact of Discrete wavelet transform - balance sparsity norms (DWT-BSN) denoising is seen in Table-7 and Fig. 11 above, with particular focus on the metrics of STOI and PESQ across different wavelet families. With a highest PESQ improvement of 4.1626, the obtained results show that the Haar wavelet significantly improves speech quality. Additionally, the denoised signal which has highest STOI value of 0.9687 was obtained from Symlet-15 wavelet. All the results show that for every wavelet has a clear increase in STOI for denoised signal.
4.3 DWT-FFU
Using Discrete wavelet transform - Fixed-Form Universal (DWT-FFU) method for suppressing the background noise, Table-8 and Fig. 12–13 displays the denoising results obtained for various wavelet types. The results shows the selected wavelet families improved MSE and SNR metrics for 5 dB noisy signal. The Symlet-15 wavelet raises SNR by 13.2295 db, suggesting that DWT-FFU thresholding greatly enhances audio signal quality for hearing aids.
Table-8: Assessing the Effectiveness of the Discrete wavelet transform - Fixed-Form Universal (DWT-FFU) Denoising Technique Using Various Wavelets
Wavelets
|
Ini-MSE
|
Den-MSE
|
Ini-SNR (dB)
|
Den-SNR(dB)
|
haar
|
0.0016590
|
0.0007548
|
5
|
7.4705
|
db-5
|
0.0016590
|
0.0002747
|
5
|
12.5725
|
db-10
|
0.0016590
|
0.0002489
|
5
|
13.0389
|
db-15
|
0.0016590
|
0.0002450
|
5
|
13.1156
|
sym-5
|
0.0016590
|
0.0002721
|
5
|
12.6148
|
sym-10
|
0.0016590
|
0.0002474
|
5
|
13.0736
|
sym-15
|
0.0016590
|
0.0002395
|
5
|
13.2295
|
coif-3
|
0.0016590
|
0.0002590
|
5
|
12.8533
|
coif-4
|
0.0016590
|
0.0002500
|
5
|
13.0232
|
coif-5
|
0.0016590
|
0.0002449
|
5
|
13.1222
|
bior-5.5
|
0.0016590
|
0.0003515
|
5
|
11.3489
|
The denoising results obtained for various wavelet families from Discrete wavelet transform - Fixed-Form Universal (DWT-FFU) in terms of STOI and PESQ are displayed in Table-9 and the related Fig. 14. The denoised results show improvement in Speech quality and highest PESQ value improvement obtained is 2.7405 for Haar wavelet and wavelet Symlet-15 has the highest STOI of 0.8649 after denoising noisy speech signals using DWT-Fixed-Form Universal (DWT-FFU) threshold.
Table-9: Trends in Discrete wavelet transform - Fixed-Form Universal (DWT-FFU) Denoising using speech quality Metrics
Wavelets
|
Ini-STOI
|
Den-STOI
|
Ini-PESQ
|
Den-PESQ
|
haar
|
0.5167
|
0.6545
|
1.3067
|
2.7405
|
db-5
|
0.5167
|
0.8089
|
1.3067
|
2.1610
|
db-10
|
0.5167
|
0.8415
|
1.3067
|
1.7599
|
db-15
|
0.5167
|
0.8602
|
1.3067
|
1.6080
|
sym-5
|
0.5167
|
0.8098
|
1.3067
|
2.1453
|
sym-10
|
0.5167
|
0.8481
|
1.3067
|
1.7567
|
sym-15
|
0.5167
|
0.8649
|
1.3067
|
1.6345
|
coif-3
|
0.5167
|
0.8227
|
1.3067
|
2.0160
|
coif-4
|
0.5167
|
0.8349
|
1.3067
|
1.8463
|
coif-5
|
0.5167
|
0.8419
|
1.3067
|
1.7496
|
bior-5.5
|
0.5167
|
0.8149
|
1.3067
|
1.8145
|
4.4 SS Speech Enhancement
Denoising performance of Spectral Subtraction algorithm is insightfully analyzed for hearing aids in Table-10 and Figs. 15–16 for speech noisy signals with SNR ranks − 5.0 dB, 00 dB, 5.0 dB, 10.0 dB, 15.0 dB and 20.0 dB. From the obtained results the filter perform well for all SNR levels and gave better STOI, higher PESQ, higher SNR, and lower MSE for reducing background noise in Speech signal. Even in challenging situations, such as when the − 5 dB SNR, this filter enhanced speech quality to 18.67631 dB SNR value. The proposed SS method greatly enhance quality as well as intelligibility of processed audio signals while effectively decreasing noise. For higher noise situation like − 5 dB noisy signal it gave 0.99278 stoi and 4.29640 PESQ score improvements.
Table-10: Evaluation metrics obtained of audio signal with multiple SNR ranks during evaluation of SS denoising
Audio Signal
|
Ini-MSE
|
Den-MSE
|
Ini-SNR (dB)
|
Den-SNR (dB)
|
Ini-STOI
|
Den-STOI
|
Ini-PESQ
|
Den-PESQ
|
–5.0 db
|
0.0154376
|
0.0001523
|
-5.00000
|
18.67631
|
0.00727
|
0.99278
|
0.11432
|
4.29640
|
0.0 db
|
0.0052356
|
0.0000788
|
0.00000
|
20.77442
|
0.41326
|
0.99601
|
0.85745
|
4.33310
|
5.0 db
|
0.0016590
|
0.0000343
|
5.00000
|
22.26100
|
0.51673
|
0.99414
|
1.30670
|
4.25440
|
10.0 db
|
0.000527
|
0.0000240
|
10.00000
|
23.31341
|
0.65305
|
0.98980
|
1.93012
|
4.15170
|
15.0 db
|
0.0001647
|
0.0000122
|
15.00000
|
25.22735
|
0.72535
|
0.98215
|
2.40260
|
3.91690
|
20.0 dB
|
0.0000537
|
0.0000136
|
20.00000
|
24.99511
|
0.79457
|
0.96641
|
2.91465
|
3.60430
|
4.5 SS-DWTBSN
The Table-11 and Figs. 17–18 illustrate the outcomes after applying proposed SS-DWTBSN Speech Enhancement method for hearing aids. Experimental results show that this proposed method gave higher SNR improvement for all wavelets in environmental background noise of 5 dB SNR. There is a substantial reduction in MSE, with values as low as 0.00002316, and a notable-increase in SNR of 23.5983 dB for Symlet-15 wavelet. These results underscore the efficacy of SS-DWTBSN in significantly reducing noise and improving audio quality, with sym15 wavelet delivering exceptional performance.
Table-11: Evaluation of MSE and SNR Before and After SS-DWTBSN Denoising Across Wavelet Families
Wavelet
|
Ini_MSE
|
Den_MSE
|
Ini_SNR (dB)
|
Den_SNR (dB)
|
haar
|
0.0016590
|
0.00002581
|
5
|
23.0983
|
db5
|
0.0016590
|
0.00002331
|
5
|
23.5693
|
db10
|
0.0016590
|
0.00002322
|
5
|
23.5872
|
db15
|
0.0016590
|
0.00002324
|
5
|
23.5839
|
sym5
|
0.0016590
|
0.00002333
|
5
|
23.5653
|
sym10
|
0.0016590
|
0.00002328
|
5
|
23.5754
|
sym15
|
0.0016590
|
0.00002316
|
5
|
23.5983
|
coif3
|
0.0016590
|
0.00002331
|
5
|
23.5688
|
coif4
|
0.0016590
|
0.00002327
|
5
|
23.5772
|
coif5
|
0.0016590
|
0.00002325
|
5
|
23.5815
|
bior5.5
|
0.0016590
|
0.00002373
|
5
|
23.4858
|
The Below Table-12 and Fig. 19 contain a comparison of STOI and PESQ metrics for initial noisy signal and denoised signal. SS-DWTBSN Speech Enhancement method has a significant improvement in both STOI and PESQ metrics across all wavelets, reflecting enhanced speech intelligibility and quality. The highest improvement achieved in STOI is 0.9965 for coif5 wavelet and PESQ is 4.3469 for db5 wavelet. Generally, all wavelet families validate substantial improvements in speech quality and intelligibility after denoising.
Table-12: STOI and PESQ Metrics Pre and Post SS-DWTBSN Denoising Through Wavelet Families
Wavelet
|
Ini_STOI
|
Den_STOI
|
Ini_PESQ
|
Den_PESQ
|
haar
|
0.5167
|
0.9951
|
1.3067
|
4.3383
|
db5
|
0.5167
|
0.9964
|
1.3067
|
4.3469
|
db10
|
0.5167
|
0.9964
|
1.3067
|
4.3364
|
db15
|
0.5167
|
0.9965
|
1.3067
|
4.3418
|
sym5
|
0.5167
|
0.9964
|
1.3067
|
4.3439
|
sym10
|
0.5167
|
0.9964
|
1.3067
|
4.3480
|
sym15
|
0.5167
|
0.9965
|
1.3067
|
4.3427
|
coif3
|
0.5167
|
0.9964
|
1.3067
|
4.3535
|
coif4
|
0.5167
|
0.9964
|
1.3067
|
4.3532
|
coif5
|
0.5167
|
0.9965
|
1.3067
|
4.3530
|
bior5.5
|
0.5167
|
0.9965
|
1.3067
|
4.3368
|
4.6 SS-DWTFFU
The experimental results obtained from proposed SS-DWTFFU are given in Table-13 and accompanying Figs. 19–20 for various wavelet families. Among the wavelets, the Symlet-15 wavelet stands out as it achieves the best MSE value of 0.00007678 and SNR value of 18.4446 dB for denoised speech signal, indicating effective noise reduction for hearing aids.
Table-13: Denoising Results for SS-DWTFFU with Various Wavelet Families
Wavelet
|
Ini_MSE
|
Den_MSE
|
Ini_SNR (dB)
|
Den_SNR (dB)
|
haar
|
0.0016590
|
0.00021413
|
5
|
13.8315
|
db5
|
0.0016590
|
0.00008315
|
5
|
18.0883
|
db10
|
0.0016590
|
0.00007828
|
5
|
18.3578
|
db15
|
0.0016590
|
0.00007894
|
5
|
18.3232
|
sym5
|
0.0016590
|
0.00008306
|
5
|
18.0943
|
sym10
|
0.0016590
|
0.00007873
|
5
|
18.3336
|
sym15
|
0.0016590
|
0.00007678
|
5
|
18.4446
|
coif3
|
0.0016590
|
0.00008081
|
5
|
18.2161
|
coif4
|
0.0016590
|
0.00007927
|
5
|
18.3023
|
coif5
|
0.0016590
|
0.00007815
|
5
|
18.3664
|
bior5.5
|
0.0016590
|
0.00008049
|
5
|
18.2284
|
Table-14 and the accompanying Fig. 21 contain the noise reduction results in terms of STOI and PESQ for different wavelet families when SS-DWTFFU Speech Enhancement method applied for hearing aids. The experimental results show that using proposed method for haar wavelet highest PESQ values of 3.5576 achieved and notably showing a great improvement in speech quality. In addition, highest STOI of 0.9214 achieved by Symlet-15 wavelet.
Table-14: Initial and denoised STOI and PESQ Metrics for SS-DWTFFU Denoising
Wavelets
|
Ini_STOI
|
Den_STOI
|
Ini_PESQ
|
Den_PESQ
|
haar
|
0.5167
|
0.8604
|
1.3067
|
3.5576
|
db5
|
0.5167
|
0.9054
|
1.3067
|
3.2225
|
db10
|
0.5167
|
0.9176
|
1.3067
|
2.5566
|
db15
|
0.5167
|
0.9181
|
1.3067
|
2.3089
|
sym5
|
0.5167
|
0.9070
|
1.3067
|
3.1893
|
sym10
|
0.5167
|
0.9148
|
1.3067
|
2.5335
|
sym15
|
0.5167
|
0.9214
|
1.3067
|
2.3233
|
coif3
|
0.5167
|
0.9095
|
1.3067
|
2.9684
|
coif4
|
0.5167
|
0.9147
|
1.3067
|
2.6747
|
coif5
|
0.5167
|
0.9142
|
1.3067
|
2.4946
|
4.7 Comparing performance of Speech Enhancement algorithms
Comparing the suggested Speech Enhancement algorithms for hearing aids with earlier methodologies, in Table-15, we observe our methods for a noisy speech signal of low SNR of 5 dB and 10 dB. Our proposed method, SS-DWTBSN (Sym15), succeeds with the highest Signal-to-Noise Ratio (SNR) of 23.5983 for environmental noise of 5 dB SNR in a speech signal compared to all other Speech Enhancement Techniques. The SNR comparison hints that proposed speech enhancement performance is able to give high SNR values in hearing aids. Furthermore, all our proposed methods DWT-BSN (sym15), DWTFFU (sym15), SS-DWTBSN (Sym15), SS-DWTFFU (sym15), and SS reveal major improvements in SNR compared to past techniques.
Table-15: Evaluation of results achieved from speech enhancement strategies on using various background noises
From: A Comparative Analysis of Proposed DWT and SS filters with Conventional Speech Enhancement Techniques
Reference Algorithms
|
Type of Noise
|
Audio signal in HZ
|
Outputs
|
Initial-SNRs (dB)
|
Final-SNRs (dB)
|
Wavelet along detail coefficients [46]
|
Noise-x92
|
16K
|
− 5.0
|
6.040
|
0.0
|
7.340
|
5.0
|
7.860
|
Wavelet-Wiener filter along with detail coefficients [46]
|
Noise-x92
|
16K
|
-5.0
|
8.480
|
0.0
|
9.220
|
5.0
|
9.910
|
DWT-thresholdings [47]
|
AWGN-Restaurant noise and Car noise
|
8K
|
5.0
|
10.190
|
5.0
|
7.510
|
5.0
|
8.440
|
WTD-LMS [48]
|
Aircraft Engine Noise
|
8K
|
0.0
|
19.930
|
5.0
|
21.360
|
LMSSS [43]
|
MUSIC NOISE
|
--
|
0.0
|
13.96220
|
5.0
|
15.67160
|
10.0
|
16.46080
|
MODWT (coif5)[49]
|
Sqtwolog
|
NOIZEUS
|
--
|
10.0
|
8.7329
|
Minimax
|
10.0
|
9.4118
|
Rigrsure
|
10.0
|
10.4678
|
Proposed DWT-BSN (sym_15)
|
Wham
|
8K
|
5.0
|
21.5293
|
10.0
|
21.97586
|
Proposed DWT-FFU (sym_15)
|
Wham
|
8K
|
5.0
|
13.2295
|
10.0
|
16.07837
|
Proposed SS for different SNR levels
|
Wham
|
8K
|
-5.0
|
18.67631
|
0.0
|
20.77442
|
5.0
|
22.26100
|
10.0
|
23.31341
|
25.0
|
25.22735
|
20.0
|
24.99511
|
Proposed SS-DWTBSN (Sym15)
|
Wham
|
8k
|
5.0
|
23.5983
|
10.0
|
23.96527
|
Proposed SS-DWTFFU (Sym15)
|
Wham
|
8k
|
5.0
|
18.4446
|
10.0
|
21.27231
|