An unsupervised approach for improving speech enhancement using wavelet packet transform and adaptive thresholding
Saved in:
| Main Author: | |
|---|---|
| Format: | Artículo científico |
| Language: | en |
| Published: |
Universidad de Carabobo
2019
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866814421274198016 |
|---|---|
| author | Mohammadali Shafieian |
| author_facet | Mohammadali Shafieian |
| contents | An unsupervised approach for improving speech enhancement using wavelet packet transform and adaptive thresholding Mohammadali Shafieian Mojdeh Rahmanian Ingeniería speech denoising adaptive threshold spectral subtraction Wavelet packet analysis voice activity detection In this article is proposed a method for improving speech enhancement techniques that use wavelet packet transform by applying adaptive thresholds on wavelet packet coefficients and using voice activity detection as well as applying spectral subtraction technique. The adaptive thresholds are determined according to the level of noise in the noisy speech signal. Furthermore, principal component analysis method is used as a powerful statistical method and linear transform technique in analyzing wavelet packet coefficients. An advantage of the proposed methods is that unlike other algorithms based on wavelet packet transform in which detection of unvoiced part of speech signal affects the performance of the algorithms considerably, proposed methods don’t require any tool to detect voice or unvoiced part of speech signal. The voice activity detection utilized is able to update noise statistics which is beneficial for the colored and non-stationary noises. The proposed methods were evaluated for speech signals containing 30 sentences in NOIZEUS database for 5 different noise types. Simulation results show that using wavelet packet transform combined with adaptive thresholding in our proposed methods outperform similar methods and can significantly enhance the quality of noisy speech for different types of noises. Eventually, evaluation of performance criteria such as SDR, SAR, SIR and SegSNR confirm the ability of the method for speech enhancement. 2019 artículo científico 1316-6832 https://www.redalyc.org/articulo.oa?id=70762652012 https://www.redalyc.org/journal/707/70762652012/ https://www.redalyc.org/journal/707/70762652012/html/ https://www.redalyc.org/journal/707/70762652012/70762652012.epub https://www.redalyc.org/journal/707/70762652012/movil en http://www.redalyc.org/revista.oa?id=707 Revista INGENIERÍA UC application/pdf Universidad de Carabobo Revista INGENIERÍA UC (República Bolivariana de Venezuela) Num.3 Vol.26 |
| format | Artículo científico |
| id | redalyc_70762652012 |
| language | en |
| publishDate | 2019 |
| publisher | Universidad de Carabobo |
| spellingShingle | An unsupervised approach for improving speech enhancement using wavelet packet transform and adaptive thresholding Mohammadali Shafieian Ingeniería speech denoising adaptive threshold spectral subtraction Wavelet packet analysis voice activity detection An unsupervised approach for improving speech enhancement using wavelet packet transform and adaptive thresholding Mohammadali Shafieian Mojdeh Rahmanian Ingeniería speech denoising adaptive threshold spectral subtraction Wavelet packet analysis voice activity detection In this article is proposed a method for improving speech enhancement techniques that use wavelet packet transform by applying adaptive thresholds on wavelet packet coefficients and using voice activity detection as well as applying spectral subtraction technique. The adaptive thresholds are determined according to the level of noise in the noisy speech signal. Furthermore, principal component analysis method is used as a powerful statistical method and linear transform technique in analyzing wavelet packet coefficients. An advantage of the proposed methods is that unlike other algorithms based on wavelet packet transform in which detection of unvoiced part of speech signal affects the performance of the algorithms considerably, proposed methods don’t require any tool to detect voice or unvoiced part of speech signal. The voice activity detection utilized is able to update noise statistics which is beneficial for the colored and non-stationary noises. The proposed methods were evaluated for speech signals containing 30 sentences in NOIZEUS database for 5 different noise types. Simulation results show that using wavelet packet transform combined with adaptive thresholding in our proposed methods outperform similar methods and can significantly enhance the quality of noisy speech for different types of noises. Eventually, evaluation of performance criteria such as SDR, SAR, SIR and SegSNR confirm the ability of the method for speech enhancement. 2019 artículo científico 1316-6832 https://www.redalyc.org/articulo.oa?id=70762652012 https://www.redalyc.org/journal/707/70762652012/ https://www.redalyc.org/journal/707/70762652012/html/ https://www.redalyc.org/journal/707/70762652012/70762652012.epub https://www.redalyc.org/journal/707/70762652012/movil en http://www.redalyc.org/revista.oa?id=707 Revista INGENIERÍA UC application/pdf Universidad de Carabobo Revista INGENIERÍA UC (República Bolivariana de Venezuela) Num.3 Vol.26 |
| title | An unsupervised approach for improving speech enhancement using wavelet packet transform and adaptive thresholding |
| topic | Ingeniería speech denoising adaptive threshold spectral subtraction Wavelet packet analysis voice activity detection |
| url | https://www.redalyc.org/articulo.oa?id=70762652012 https://www.redalyc.org/journal/707/70762652012/ https://www.redalyc.org/journal/707/70762652012/html/ https://www.redalyc.org/journal/707/70762652012/70762652012.epub https://www.redalyc.org/journal/707/70762652012/movil |