Spectrally shaped steady noise is commonly used as a masker of speech. The effects of inherent random fluctuations in amplitude of such a noise are typically ignored. Here, the importance of these random fluctuations was assessed by comparing two cases. For one, speech was mixed with steady speech-shaped noise and N-channel tone vocoded, a process referred to as signal-domain mixing (SDM); this preserved the random fluctuations of the noise. For the second, the envelope of speech alone was extracted for each vocoder channel and a constant was added corresponding to the root-mean-square value of the noise envelope for that channel. This is referred to as envelope-domain mixing (EDM); it removed the random fluctuations of the noise. Sinusoidally modulated noise and a single talker were also used as backgrounds, with both SDM and EDM. Speech intelligibility was measured for N = 12, 19, and 30, with the target-to-background ratio fixed at -7 dB. For SDM, performance was best for the speech background and worst for the steady noise. For EDM, this pattern was reversed. Intelligibility with steady noise was consistently very poor for SDM, but near-ceiling for EDM, demonstrating that the random fluctuations in steady noise have a large effect.

Objective:Hearing loss at high frequencies produces perceptual difficulties and is often an early sign of a more general hearing loss. This study reports the development and validation of two new speech-based hearing screening tests in English that focus on detecting hearing loss at frequencies above 2000 Hz.Design:The Internet-delivered, speech-in noise tests used closed target-word sets of digit triplets or consonant–vowel–consonant (CVC) words presented against a speech-shaped noise masker. The digit triplet test uses the digits 0 to 9 (excluding the disyllabic 7), grouped in quasi-random triplets. The CVC test uses simple words (e.g., “cat”) selected for the high-frequency spectral content of the consonants. During testing, triplets or CVC words were identified in an adaptive procedure to obtain the speech reception threshold (SRT) in noise. For these new, high-frequency (HF) tests, the noise was low-pass filtered to produce greater masking of the low-frequency speech components, increasing the sensitivity of the test for HF hearing loss. Individual test tokens (digits, CVCs) were first homogenized using a group of 10 normal-hearing (NH) listeners by equalizing intelligibility across tokens at several speech-in-noise levels. Both tests were then validated and standardized using groups of 24 NH listeners and 50 listeners with hearing impairment. Performance on the new high frequency digit triplet (HF-triplet) and CVC (HF-CVC) tests was compared with audiometric hearing loss, and with that on the unfiltered, broadband digit triplet test (BB-triplet) test, and the ASL (Adaptive Sentence Lists) speech-in-noise test.Results:The HF-triplet and HF-CVC test results (SRT) both correlated positively and highly with high-frequency audiometric hearing loss and with the ASL test. SRT for both tests as a function of high-frequency hearing loss increased at nearly three times the rate as that of the BB-triplet test. The intraindividual variability (SD) on the tests was about 2.1 (HF-triplet) and 1.7 (HF-CVC) times less than that for the BB-triplet test. The effect on the HF-triplet test of varying presentation method (professional or cheap headphones and loudspeakers) was small for the NH group and somewhat larger, but nonsignificant for the hearing-impaired group. Test repetition produced a moderate, significant learning effect for the first and second retests, but was small and nonsignificant for further retesting. The learning effect was about two times larger for the HF-CVC test than for the HF-triplet test. The sensitivity of both new tests for high-frequency hearing loss was similar, with an 87% true-positive and 7% false-positive ratio for detecting an average high-frequency hearing loss of 20 dB or more.Conclusions:The new HF-triplet and HF-CVC tests provide a sensitive and accurate method for detecting high-frequency hearing loss. The tests may signal developing hearing impairment at an early stage. The HF-triplet is preferred over the HF-CVC test because of its smaller learning effect, smaller error rate, greater ...

Tinnitus has been linked to noise exposure, a common form of which is listening to music as a leisure activity. The relationship between tinnitus and type and duration of music exposure is not well understood. We conducted an internet-based population study that asked participants questions about lifetime music exposure and hearing, and included a hearing test involving speech intelligibility in noise, the High Frequency Digit Triplets Test. 4950 people aged 17–75 years completed all questions and the hearing test. Results were analyzed using multinomial regression models. High exposure to leisure music, hearing difficulty, increasing age and workplace noise exposure were independently associated with increased tinnitus. Three forms of music exposure (pubs/clubs, concerts, personal music players) did not differ in their relationship to tinnitus. More males than females reported tinnitus. The objective measure of speech reception threshold had only a minimal relationship with tinnitus. Self-reported hearing difficulty was more strongly associated with tinnitus, but 76% of people reporting usual or constant tinnitus also reported little or no hearing difficulty. Overall, around 40% of participants of all ages reported never experiencing tinnitus, while 29% reported sometimes, usually or constantly experiencing tinnitus that lasted more than 5 min. Together, the results suggest that tinnitus is much more common than hearing loss, but that there is little association between the two, especially among the younger adults disproportionately sampled in this study.

OBJECTIVE: To use machine learning in the form of a deep neural network to objectively classify paired auditory brainstem response waveforms into either: 'clear response', 'inconclusive' or 'response absent'. DESIGN: A deep convolutional neural network was trained and fine-tuned using stratified 10-fold cross-validation on 190 paired ABR waveforms. The final model was evaluated on a test set of 42 paired waveforms. STUDY SAMPLE: The full dataset comprised 232 paired ABR waveforms recorded from eight normal-hearing individuals. The dataset was obtained from the PhysioBank database. The paired waveforms were independently labelled by two audiological scientists in order to train and evaluate the network's performance. RESULTS: The trained neural network was able to classify paired ABR waveforms with 92.9% accuracy. The sensitivity and specificity were 92.9% and 96.4% respectively. CONCLUSIONS: This neural network may have clinical utility in assisting clinicians with waveform classification for the purpose of hearing threshold estimation. Further evaluation on a large clinically-obtained dataset would provide further validation with regards to the clinical potential of the neural network in diagnostic adult testing, newborn testing and in automated newborn hearing screening.