Limits...
Effects of Adaptation Rate and Noise Suppression on the Intelligibility of Compressed-Envelope Based Speech.

Lai YH, Tsao Y, Chen F - PLoS ONE (2015)

Bottom Line: Experimental results demonstrated that the adaptation rate in envelope compression had a notable effect on the speech intelligibility performance of the AEC strategy.By specifying a suitable adaptation rate, speech intelligibility could be enhanced significantly in noise compared to when using static envelope compression.Moreover, results confirmed that the AEC strategy was suitable for combining with noise reduction to improve the intelligibility of envelope-based speech in noise.

View Article: PubMed Central - PubMed

Affiliation: Research Center for Information Technology Innovation, Academia Sinica, Taipei, Taiwan.

ABSTRACT
Temporal envelope is the primary acoustic cue used in most cochlear implant (CI) speech processors to elicit speech perception for patients fitted with CI devices. Envelope compression narrows down envelope dynamic range and accordingly degrades speech understanding abilities of CI users, especially under challenging listening conditions (e.g., in noise). A new adaptive envelope compression (AEC) strategy was proposed recently, which in contrast to the traditional static envelope compression, is effective at enhancing the modulation depth of envelope waveform by making best use of its dynamic range and thus improving the intelligibility of envelope-based speech. The present study further explored the effect of adaptation rate in envelope compression on the intelligibility of compressed-envelope based speech. Moreover, since noise reduction is another essential unit in modern CI systems, the compatibility of AEC and noise reduction was also investigated. In this study, listening experiments were carried out by presenting vocoded sentences to normal hearing listeners for recognition. Experimental results demonstrated that the adaptation rate in envelope compression had a notable effect on the speech intelligibility performance of the AEC strategy. By specifying a suitable adaptation rate, speech intelligibility could be enhanced significantly in noise compared to when using static envelope compression. Moreover, results confirmed that the AEC strategy was suitable for combining with noise reduction to improve the intelligibility of envelope-based speech in noise.

No MeSH data available.


Related in: MedlinePlus

Examples of amplitude envelope processed by (a) SEC, (b) AEC with a very slow step size (i.e., Δα = 0.0001), and (c) AEC with a fast step size (i.e., Δα = 0.1).The envelope was extracted from the 6th channel of a testing sentence masked by SSN at 5 dB, and compressed to 5 dB DR.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4510405&req=5

pone.0133519.g008: Examples of amplitude envelope processed by (a) SEC, (b) AEC with a very slow step size (i.e., Δα = 0.0001), and (c) AEC with a fast step size (i.e., Δα = 0.1).The envelope was extracted from the 6th channel of a testing sentence masked by SSN at 5 dB, and compressed to 5 dB DR.

Mentions: The results of Experiment 1 further indicate that the value of adaptation rate Δα affects the intelligibility of AEC-processed speech. As noted, the SF is an important parameter in the AEC strategy. The use of an inappropriate value of Δα may reduce the speech intelligibility benefits, especially in test conditions with interfering masker (i.e., 2T masker in this study). When a very small value of Δα is used in the AEC strategy, it is difficult to obtain a larger SF for short-term signals, thereby also making it difficult to obtain a larger modulation depth from AEC. On the other hand, the use of a very large value of Δα may induce the perception of pumping and increase the probability of the processed envelope falling within the range of peak clipping and thereby causing distortion. Therefore, the value of adaptation rate needs to be selected as a trade-off between speech intelligibility and the pumping effect and distortion. Fig 8 exemplifies results obtained by traditional SEC, AEC with a very slow step size (i.e., Δα = 0.0001), and AEC with a fast step size (i.e., Δα = 0.1). This example shows that using a very small value of Δα [0.0001 in Fig 8(b)] limits the benefit of AEC processing, because the step size is too small over a short time period to provide a sufficiently large modulation depth relative to the traditional SEC strategy. In contrast, when the value of Δα is set too large [0.1 in Fig 8(c)], the step size is too large over a short time period, which will result in a considerable amount of envelope information appearing outside of the DR and thus cause distortion by peak clipping and decrease the intelligibility of AEC-processed speech. In addition, when the envelope waveform varies too rapidly in the AEC strategy, it will also cause the perception of pumping to affect the performance of speech intelligibility. The results of Experiment 1 indicate that, on average, Δα = 0.001 provides a higher speech intelligibility score than the other values (i.e., Δα = 0.01 and 0.1), especially in test conditions with 2T masker. Therefore, Δα = 0.001 strikes the optimal balance among benefits, pumping effect, and distortion in the AEC strategy. Since we only used these three values of Δα (i.e., slow, moderate, and fast step size) to investigate the performance of the AEC strategy in this study, future studies should further investigate its effect while considering the wider characteristics of language (e.g., the band importance function or tone of Mandarin) and noise types. In addition, the technology of machine learning could be used to help tune this parameter in future studies.


Effects of Adaptation Rate and Noise Suppression on the Intelligibility of Compressed-Envelope Based Speech.

Lai YH, Tsao Y, Chen F - PLoS ONE (2015)

Examples of amplitude envelope processed by (a) SEC, (b) AEC with a very slow step size (i.e., Δα = 0.0001), and (c) AEC with a fast step size (i.e., Δα = 0.1).The envelope was extracted from the 6th channel of a testing sentence masked by SSN at 5 dB, and compressed to 5 dB DR.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4510405&req=5

pone.0133519.g008: Examples of amplitude envelope processed by (a) SEC, (b) AEC with a very slow step size (i.e., Δα = 0.0001), and (c) AEC with a fast step size (i.e., Δα = 0.1).The envelope was extracted from the 6th channel of a testing sentence masked by SSN at 5 dB, and compressed to 5 dB DR.
Mentions: The results of Experiment 1 further indicate that the value of adaptation rate Δα affects the intelligibility of AEC-processed speech. As noted, the SF is an important parameter in the AEC strategy. The use of an inappropriate value of Δα may reduce the speech intelligibility benefits, especially in test conditions with interfering masker (i.e., 2T masker in this study). When a very small value of Δα is used in the AEC strategy, it is difficult to obtain a larger SF for short-term signals, thereby also making it difficult to obtain a larger modulation depth from AEC. On the other hand, the use of a very large value of Δα may induce the perception of pumping and increase the probability of the processed envelope falling within the range of peak clipping and thereby causing distortion. Therefore, the value of adaptation rate needs to be selected as a trade-off between speech intelligibility and the pumping effect and distortion. Fig 8 exemplifies results obtained by traditional SEC, AEC with a very slow step size (i.e., Δα = 0.0001), and AEC with a fast step size (i.e., Δα = 0.1). This example shows that using a very small value of Δα [0.0001 in Fig 8(b)] limits the benefit of AEC processing, because the step size is too small over a short time period to provide a sufficiently large modulation depth relative to the traditional SEC strategy. In contrast, when the value of Δα is set too large [0.1 in Fig 8(c)], the step size is too large over a short time period, which will result in a considerable amount of envelope information appearing outside of the DR and thus cause distortion by peak clipping and decrease the intelligibility of AEC-processed speech. In addition, when the envelope waveform varies too rapidly in the AEC strategy, it will also cause the perception of pumping to affect the performance of speech intelligibility. The results of Experiment 1 indicate that, on average, Δα = 0.001 provides a higher speech intelligibility score than the other values (i.e., Δα = 0.01 and 0.1), especially in test conditions with 2T masker. Therefore, Δα = 0.001 strikes the optimal balance among benefits, pumping effect, and distortion in the AEC strategy. Since we only used these three values of Δα (i.e., slow, moderate, and fast step size) to investigate the performance of the AEC strategy in this study, future studies should further investigate its effect while considering the wider characteristics of language (e.g., the band importance function or tone of Mandarin) and noise types. In addition, the technology of machine learning could be used to help tune this parameter in future studies.

Bottom Line: Experimental results demonstrated that the adaptation rate in envelope compression had a notable effect on the speech intelligibility performance of the AEC strategy.By specifying a suitable adaptation rate, speech intelligibility could be enhanced significantly in noise compared to when using static envelope compression.Moreover, results confirmed that the AEC strategy was suitable for combining with noise reduction to improve the intelligibility of envelope-based speech in noise.

View Article: PubMed Central - PubMed

Affiliation: Research Center for Information Technology Innovation, Academia Sinica, Taipei, Taiwan.

ABSTRACT
Temporal envelope is the primary acoustic cue used in most cochlear implant (CI) speech processors to elicit speech perception for patients fitted with CI devices. Envelope compression narrows down envelope dynamic range and accordingly degrades speech understanding abilities of CI users, especially under challenging listening conditions (e.g., in noise). A new adaptive envelope compression (AEC) strategy was proposed recently, which in contrast to the traditional static envelope compression, is effective at enhancing the modulation depth of envelope waveform by making best use of its dynamic range and thus improving the intelligibility of envelope-based speech. The present study further explored the effect of adaptation rate in envelope compression on the intelligibility of compressed-envelope based speech. Moreover, since noise reduction is another essential unit in modern CI systems, the compatibility of AEC and noise reduction was also investigated. In this study, listening experiments were carried out by presenting vocoded sentences to normal hearing listeners for recognition. Experimental results demonstrated that the adaptation rate in envelope compression had a notable effect on the speech intelligibility performance of the AEC strategy. By specifying a suitable adaptation rate, speech intelligibility could be enhanced significantly in noise compared to when using static envelope compression. Moreover, results confirmed that the AEC strategy was suitable for combining with noise reduction to improve the intelligibility of envelope-based speech in noise.

No MeSH data available.


Related in: MedlinePlus