Speech Intelligibility Evaluation for Mobile Phones

Jørgensen, S.; Cubick, Jens; Dau, Torsten

doi:10.3813/aaa.918896

Cited by 16 publications

(13 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This algorithm predicts intelligibility from clean and degraded speech by averaging the sample correlation coefficient of short segments of clean and degraded envelopes across 15 one-third octave bands. The STOI measure has been shown to correlate well with measured intelligibility in conditions including different additive noise sources [13], noise reduction processing [42], [13], hearing-aid and cochlear implant processing [25], and noisy speech transmitted via telephone [43]. Later work has shown the STOI measure to be closely related to an estimate of information transmission [32].…”

Section: Introductionmentioning

confidence: 95%

Nonintrusive Speech Intelligibility Prediction Using Convolutional Neural Networks

Andersen

Haan

Jensen

2018

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

Speech Intelligibility Prediction (SIP) algorithms are becoming popular tools within the development and operation of speech processing devices and algorithms. However, many SIP algorithms require knowledge of the underlying clean speech; a signal that is often not available in real-world applications. This has led to increased interest in non-intrusive SIP algorithms, which do not require clean speech to make predictions. In this paper we investigate the use of Convolutional Neural Networks (CNNs) for non-intrusive SIP. To do so, we utilize a CNN architecture that shows similarities to existing SIP algorithms, in terms of computational structure, and which allows for easy and meaningful visualization and interpretation of trained weights. We evaluate this architecture using a large dataset obtained by combining datasets from the literature. The proposed method shows high prediction performance when compared with four existing intrusive and non-intrusive SIP algorithms. This demonstrates the potential of deep learning for speech intelligibility prediction.

show abstract

Section: Introductionmentioning

confidence: 95%

Nonintrusive Speech Intelligibility Prediction Using Convolutional Neural Networks

Andersen

Haan

Jensen

2018

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

show abstract

“…This observation is confirmed for hearing impaired listeners in [28], which shows that the CSII and the STOI measure perform favorably to other measures at predicting the effect of noise reduction algorithms. The STOI measure has later been shown to compare well with other measures with respect to predicting the impact of a number of detrimental effects and processing schemes relevant to users of hearing aids and cochlear implants [8] and for predicting the intelligibility of noisy speech transmitted by telephone [4]. Finally, we mention the Speech Intelligibility prediction based on Mutual Information (SIMI) measure [12], which is very similar to the STOI measure in structure and performance, but which is based on information theoretical considerations.…”

mentioning

confidence: 99%

Predicting the Intelligibility of Noisy and Nonlinearly Processed Binaural Speech

Andersen

Haan

Jensen

2016

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

Objective speech intelligibility measures are gaining popularity in the development of speech enhancement algorithms and speech processing devices such as hearing aids. Such devices may process the input signals non-linearly and modify the binaural cues presented to the user. We propose a method for predicting the intelligibility of noisy and non-linearly processed binaural speech. This prediction is based on the noisy and processed signal as well as a clean speech reference signal. The method is obtained by extending a modified version of the short-time objective intelligibility (STOI) measure with a modified equalization-cancellation (EC) stage. We evaluate the performance of the method by comparing the predictions with measured intelligibility from four listening experiments. These comparisons indicate that the proposed measure can provide accurate predictions of 1) the intelligibility of diotic speech with an accuracy similar to that of the original STOI measure, 2) speech reception thresholds (SRTs) in conditions with a frontal target speaker and a single interferer in the horizontal plane, 3) SRTs in conditions with a frontal target and a single interferer when ideal time frequency segregation (ITFS) is applied to the left and right ears separately, and 4) the advantage of twomicrophone beamforming as applied in state-of-the-art hearing aids. A MATLAB implementation of the proposed measure is available online 1 .

show abstract

“…However, no relationship with sound pressure level and usage scenario was shown for mobile phones. For three mobile phones evaluated by Jørgensen et al (2015), transmission of speech through the mobile phones under study led to degraded speech intelligibility in comparison to the reference condition without transmission through mobile phones (Jørgensen et al, 2015). The goals of this study are: first, to determine whether behavioral aspects can have an influence on the SAR induced by a smartphone; second, to quantify speech intelligibility and in-ear sound pressure level during those same exposure scenarios; and finally, come to a conclusion about which of the three studied scenarios results in the lowest co-exposure to RF-EMF and in-ear sound pressure and whether changing from one usage scenario to another can serve as a non-technological way of reducing exposure to one or both of the studied physical agents.…”

Section: Introductionmentioning

confidence: 99%

Mobile phones: A trade-off between speech intelligibility and exposure to noise levels and to radio-frequency electromagnetic fields

Thielens

Bockstael

Declerck

et al. 2019

Environmental Research

View full text Add to dashboard Cite

When making phone calls, cellphone and smartphone users are exposed to radio-frequency (RF) electromagnetic fields (EMFs) and sound pressure simultaneously. Speech intelligibility during mobile phone calls is related to the sound pressure level of speech relative to potential background sounds and also to the RF-EMF exposure, since the signal quality is correlated with the RF-EMF strength. Additionally, speech intelligibility, sound pressure level, and exposure to RF-EMFs are dependent on how the call is made (on speaker, held at the ear, or with headsets). The relationship between speech intelligibility, sound exposure, and exposure to RF-EMFs is determined in this study. To this aim, the transmitted RF-EMF power was recorded during phone calls made by 53 subjects in three different, controlled exposure scenarios: calling with the phone at the ear, calling in speaker-mode, and calling with a headset. This emitted power is directly proportional to the exposure to RF EMFs and is translated into specific absorption rate using numerical simulations. Simultaneously, sound pressure levels have been recorded and speech intelligibility has been assessed during each phone call. The results show that exposure to RF-EMFs, quantified as the specific absorption in the head, will be reduced when speaker-mode or a headset is used, in comparison to calling next to the ear. Additionally, personal exposure to sound pressure is also found to be highest in the condition where the phone is held next to the ear. On the other hand, speech perception is found to be the best when calling with a phone next to the ear in comparison to the other studied conditions, when background noise is present. Thielens et al., 2013). These values were then renormalized to a TX power of 1 W. We chose not to model the wire of the headset in the 'Headset' configuration. Data Processing and AnalysisThe registered TX powers during each call were pooled for all participants, calls and frequencies, but split for the three exposure conditions, the two locations, and the communication technology. We chose to split the data for the two locations, since the coverage and link with the network might have been different. We chose to split the data for the different telecommunication technologies since they have distinctly different ranges of operation in terms of emitted power. We did not split the GSM results in 900 MHz and 1800 MHz. A Wilcoxon Rank Sum (WRS) test and a two-sided Kolmogorov-Smirnov (K-S) test were then used to determine whether the TX power in different exposure situations on the same test site were significantly different (significance level of 5%). In a second step, all TX powers that were recorded for GSM were multiplied by the SAR brain values at the appropriate frequency and exposure condition. Those instances where WCDMA was used to establish communication were not treated for SAR, because we did not have full information on the used frequency during those events. However, they were considered for the analysis of emitted powers. The obtained...

show abstract

Speech Intelligibility Evaluation for Mobile Phones

Cited by 16 publications

References 0 publications

Nonintrusive Speech Intelligibility Prediction Using Convolutional Neural Networks

Nonintrusive Speech Intelligibility Prediction Using Convolutional Neural Networks

Predicting the Intelligibility of Noisy and Nonlinearly Processed Binaural Speech

Mobile phones: A trade-off between speech intelligibility and exposure to noise levels and to radio-frequency electromagnetic fields

Contact Info

Product

Resources

About