Gongqiang Yu

Relationships between the modified rhyme test and objective metrics of speech intelligibility.

Relationships between the modified rhyme test (MRT) and the speech transmission index (STI), and MRT and the speech intelligibility index (SII), have been obtained for additive stationary noise and nonlinear distortion of the speech signal. The former was speech‐spectrum shaped noise, white noise, or −3‐dB/octave noise presented at speech signal‐to‐noise ratios ranging from −25 to +10 dB. The speech distortions were peak clipping or center clipping, with clipping thresholds from 2% to 98% of the cumulative magnitude histogram. Subjects (4 male, 4 female) with normal hearing were seated in an anechoic chamber. Speech was reproduced by a small, high‐fidelity loudspeaker located 2.4 m to the center‐of‐head and the inter‐aural axis. Noise was reproduced by four‐loudspeaker systems and processed to simulate a diffuse field at the ear in the horizontal plane. The revised STI was computed using the standardized test signal [IEC 60268‐16 (2003)] and the speech signal. The SII was calculated using coherence to est...

Augmented warning sound detection for hearing protectors

Perception of warning sounds, such as vehicle backup alarms, is reduced when hearing protection devices (HPDs) are worn. A cross-correlation approach is employed to detect a pre-selected warning sound and enable it to bypass the attenuation of the HPD while still attenuating the environmental noise. Computer simulation shows that the algorithm can detect the specified alarm at signal-to-environmental-noise ratios as low as -30 dB. Human subject testing of the algorithm, implemented on a modified commercial HPD, confirms the minimum detection threshold obtained in simulation, and demonstrates a 7 dB improvement in detection threshold compared with the unmodified HPD.

Understanding speech when wearing communication headsets and hearing protectors with subband processinga)

An adaptive, delayless, subband feed-forward control structure is employed to improve the speech signal-to-noise ratio (SNR) in the communication channel of a circumaural headset/hearing protector (HPD) from 90 Hz to 11.3 kHz, and to provide active noise control (ANC) from 50 to 800 Hz to complement the passive attenuation of the HPD. The task involves optimizing the speech SNR for each communication channel subband, subject to limiting the maximum sound level at the ear, maintaining a speech SNR preferred by users, and reducing large inter-band gain differences to improve speech quality. The performance of a proof-of-concept device has been evaluated in a pseudo-diffuse sound field when worn by human subjects under conditions of environmental noise and speech that do not pose a risk to hearing, and by simulation for other conditions. For the environmental noises employed in this study, subband speech SNR control combined with subband ANC produced greater improvement in word scores than subband ANC alone, and improved the consistency of word scores across subjects. The simulation employed a subject-specific linear model, and predicted that word scores are maintained in excess of 90% for sound levels outside the HPD of up to ∼115 dBA.

Modeling speech intelligibility in distortion conditions using a speech-based speech transmission index

A modified objective model expanding on the speech-based speech transmission index is proposed to predict speech intelligibility under various conditions of nonlinear distortion. The proposed model computes values over a time window by analyzing the signal to noise ratio and the modulation transfer function between the input and output of a transmission channel. The channel is divided into frequency bands with ranges compatible with critical bands. The cross covariance among adjacent frequency bands is also considered. The index is obtained by averaging the calculated values of these time windows. The proposed model is evaluated with subjective measurement of word intelligibility scores using the modified rhythm test with non-linear distortions of phase jitter and clipping introduced into the speech material presented to subjects. The results demonstrate that high correlations between the indices of the proposed model and the intelligibility scores are maintained for these distortions.

Are reports of temporary threshold shift-like symptoms by humans with normal hearing associated with hidden hearing loss?

Forty-six subjects with normal hearing, mean age 20.2 years, were selected from 451 volunteers completing a questionnaire concerning hearing, exposure to noise, experiencing TTS-like symptoms, and speech understanding. Metrics quantifying reports of TTS-like symptoms were constructed from responses to questions concerning hearing immediately after noise exposure. Statistically significant deteriorations in scores on the Speech, Spatial and Qualities of Hearing Scale (SSQ) (Gatehouse & Noble, Int J Audiol 43, 85-99 (2004)) were found with increasing values of TTS metrics for all SSQ questions. Groups reporting TTS-like symptoms (“exposed”), and “controls” (with little / no noise exposure and no reports of TTS-like symptoms), were formed from the subject pool with mean hearing levels differing <2 dB from 250Hz—8kHz. There was no difference in mean word scores between groups in a Modified Rhyme test conducted in speech-spectrum shaped noise. However, the exposed group exhibited a statistically significant de...

Method to improve warning sound detection for hearing protectors

Hearing protection devices (HPDs) provide both desirable attenuation of environmental noise and undesirable attenuation of auditory warning alarms, such as vehicle backup alarms. Both warning alarm detection and localization performance are affected. A digital cross-correlation based method is described to identify a pre-selected warning alarm to bypass the attenuation of the HPD while maintaining attenuation of environmental noise outside the bandwidth of the alarm. This method can be integrated into existing digital HPD designs. Computer simulation of the algorithm demonstrates that an alarm signal can be detected at signal-to-environmental noise ratios as low as 30 dB for the military and industrial noise sources investigated. Implementation of the method using a modified commercial HPD demonstrates a 7 dB improvement in warning alarm detection threshold compared with an unmodified HPD. Alternative methods for presenting the alarm signal to the user will be discussed as well as modifications to expand ...

Strategies for improving speech intelligibility and warning signal detection in communication headsets/hearing protectors

Strategies for improving speech understanding and warning signal detection when wearing communication headsets/hearing protectors (HPDs) in environmental noise must accommodate sounds from different sources at different times. A subband signal processing approach would appear desirable, with a delayless structure essential for active noise reduction (ANR). The requirements for communication channel and ANR controllers differ, owing to the different bandwidths required for speech and warning signals, and for ANR. Subbands for optimizing speech signal-to-noise ratios are commonly fractional-octave bandwidth, while computational efficiency favors linear subbands for ANR. Increasing the number of subbands reduces computational cost for the latter but the advantage is less apparent for communication signal control. Possibilities exist for harmonizing subband filter structures by constructing models of speech intelligibility using computationally efficient bandwidths. In contrast, algorithms for detecting warni...

Integrating speech enhancement with subband active noise control to improve communication in hearing protectors

Many workers refuse to wear hearing protection devices (HPDs) because they would rather accept the health risks than sacrifice the ability to communicate with coworkers. Integrating active noise reduction (ANR) techniques with speech enhancement algorithms could solve these limitations of modern electronic HPDs. An adaptive delay feedforward subband structure has been implemented by forming parallel signal filtering and filter update paths for each frequency band. Subband ANR provides additional attenuation of environmental noise beyond that of passive HPDs within the lower frequencies associated with speech communication. The subband structure can also provide valuable information regarding the spectral content of the environmental noise that can be exploited to determine where additional power is needed in a communication channel to improve intelligibility. The system, initially developed in simulation, maintains a specified signal-to-noise ratio in each subband while simultaneously limiting the power a...