Muddy, muddled, or muffled? Understanding the perception of audio quality in music by hearing aid users.

Introduction: Previous work on audio quality evaluation has demonstrated a developing convergence of the key perceptual attributes underlying judgments of quality, such as timbral, spatial and technical attributes. However, across existing research there remains a limited understanding of the crucial perceptual attributes that inform audio quality evaluation for people with hearing loss, and those who use hearing aids. This is especially the case with music, given the unique problems it presents in contrast to human speech. Method: This paper presents a sensory evaluation study utilising descriptive analysis methods, in which a panel of hearing aid users collaborated, through consensus, to identify the most important perceptual attributes of music audio quality and developed a series of rating scales for future listening tests. Participants (N = 12), with a hearing loss ranging from mild to severe, first completed an online elicitation task, providing single-word terms to describe the audio quality of original and processed music samples; this was completed twice by each participant, once with hearing aids, and once without. Participants were then guided in discussing these raw terms across three focus groups, in which they reduced the term space, identified important perceptual groupings of terms, and developed perceptual attributes from these groups (including rating scales and definitions for each). Results: Findings show that there were seven key perceptual dimensions underlying music audio quality (clarity, harshness, distortion, spaciousness, treble strength, middle strength, and bass strength), alongside a music audio quality attribute and possible alternative frequency balance attributes. Discussion: We outline how these perceptual attributes align with extant literature, how attribute rating instruments might be used in future work, and the importance of better understanding the music listening difficulties of people with varied profiles of hearing loss.

Improving the measurement and acoustic performance of transparent face masks and shields.

Opaque face masks harm communication by preventing speech-reading (lip-reading) and attenuating high-frequency sound. Although transparent masks and shields (visors) with clear plastic inserts allow speech-reading, they usually create more sound attenuation than opaque masks. Consequently, an iterative process was undertaken to create a better design, and the instructions to make it are published. The experiments showed that lowering the mass of the plastic inserts decreases the high-frequency sound attenuation. A shield with a clear thermoplastic polyurethane (TPU) panel had an insertion loss of (2.0 ± 1.1) dB for 1.25-8 kHz, which improves on previous designs that had attenuations of 11.9 dB and above. A cloth mask with a TPU insert was designed and had an insertion loss of (4.6 ± 2.3) dB for 2-8 kHz, which is better than the 9-22 dB reported previously in the literature. The speech intelligibility index was also evaluated. Investigations to improve measurement protocols that use either mannikins or human talkers were undertaken. Manufacturing variability and inconsistency of human speaking were greater sources of experimental error than fitting differences. It was shown that measurements from a mannikin could match those from humans if insertion losses from four human talkers were averaged.

Dataset of British English speech recordings for psychoacoustics and speech processing research: The clarity speech corpus.

This paper presents the Clarity Speech Corpus, a publicly available, forty speaker British English speech dataset. The corpus was created for the purpose of running listening tests to gauge speech intelligibility and quality in the Clarity Project, which has the goal of advancing speech signal processing by hearing aids through a series of challenges. The dataset is suitable for machine learning and other uses in speech and hearing technology, acoustics and psychoacoustics. The data comprises recordings of approximately 10,000 sentences drawn from the British National Corpus (BNC) with suitable length, words and grammatical construction for speech intelligibility testing. The collection process involved the selection of a subset of BNC sentences, the recording of these produced by 40 British English speakers, and the processing of these recordings to create individual sentence recordings with associated transcripts and metadata.

Sound Categories: Category Formation and Evidence-Based Taxonomies.

Five evidence-based taxonomies of everyday sounds frequently reported in the soundscape literature have been generated. An online sorting and category-labeling method that elicits rather than prescribes descriptive words was used. A total of N = 242 participants took part. The main categories of the soundscape taxonomy were people, nature, and manmade, with each dividing into further categories. Sounds within the nature and manmade categories, and two further individual sound sources, dogs, and engines, were explored further by repeating the procedure using multiple exemplars. By generating multidimensional spaces containing both sounds and the spontaneously generated descriptive words the procedure allows for the interpretation of the psychological dimensions along which sounds are organized. This reveals how category formation is based upon different cues - sound source-event identification, subjective-states, and explicit assessment of the acoustic signal - in different contexts. At higher levels of the taxonomy the majority of words described sound source-events. In contrast, when categorizing dog sounds a greater proportion of the words described subjective-states, and valence and arousal scores of these words correlated with their coordinates along the first two dimensions of the data. This is consistent with valence and arousal judgments being the primary categorization strategy used for dog sounds. In contrast, when categorizing engine sounds a greater proportion of the words explicitly described the acoustic signal. The coordinates of sounds along the first two dimensions were found to correlate with fluctuation strength and sharpness, consistent with explicit assessment of acoustic signal features underlying category formation for engine sounds. By eliciting descriptive words the method makes explicit the subjective meaning of these judgments based upon valence and arousal and acoustic properties, and the results demonstrate distinct strategies being spontaneously used to categorize different types of sounds.

Metadiffusers: Deep-subwavelength sound diffusers.

We present deep-subwavelength diffusing surfaces based on acoustic metamaterials, namely metadiffusers. These sound diffusers are rigidly backed slotted panels, with each slit being loaded by an array of Helmholtz resonators. Strong dispersion is produced in the slits and slow sound conditions are induced. Thus, the effective thickness of the panel is lengthened introducing its quarter wavelength resonance in the deep-subwavelength regime. By tuning the geometry of the metamaterial, the reflection coefficient of the panel can be tailored to obtain either a custom reflection phase, moderate or even perfect absorption. Using these concepts, we present ultra-thin diffusers where the geometry of the metadiffuser has been tuned to obtain surfaces with spatially dependent reflection coefficients having uniform magnitude Fourier transforms. Various designs are presented where, quadratic residue, primitive root and ternary sequence diffusers are mimicked by metadiffusers whose thickness are 1/46 to 1/20 times the design wavelength, i.e., between about a twentieth and a tenth of the thickness of traditional designs. Finally, a broadband metadiffuser panel of 3 cm thick was designed using optimization methods for frequencies ranging from 250 Hz to 2 kHz.

A metric for predicting binaural speech intelligibility in stationary noise and competing speech maskers.

One criterion in the design of binaural sound scenes in audio production is the extent to which the intended speech message is correctly understood. Object-based audio broadcasting systems have permitted sound editors to gain more access to the metadata (e.g., intensity and location) of each sound source, providing better control over speech intelligibility. The current study describes and evaluates a binaural distortion-weighted glimpse proportion metric-BiDWGP-which is motivated by better-ear glimpsing and binaural masking level differences. BiDWGP predicts intelligibility from two alternative input forms: either binaural recordings or monophonic recordings from each sound source along with their locations. Two listening experiments were performed with stationary noise and competing speech, one in the presence of a single masker, the other with multiple maskers, for a variety of spatial configurations. Overall, BiDWGP with both input forms predicts listener keyword scores with correlations of 0.95 and 0.91 for single- and multi-masker conditions, respectively. When considering masker type separately, correlations rise to 0.95 and above for both types of maskers. Predictions using the two input forms are very similar, suggesting that BiDWGP can be applied to the design of sound scenes where only individual sound sources and their locations are available.

The effect of microphone wind noise on the amplitude modulation of wind turbine noise and its mitigation.

Microphone wind noise can corrupt outdoor recordings even when wind shields are used. When monitoring wind turbine noise, microphone wind noise is almost inevitable because measurements cannot be made in still conditions. The effect of microphone wind noise on two amplitude modulation (AM) metrics is quantified in a simulation, showing that even at low wind speeds of 2.5 m/s errors of over 4 dBA can result. As microphone wind noise is intermittent, a wind noise detection algorithm is used to automatically find uncorrupted sections of the recording, and so recover the true AM metrics to within ±2/±0.5 dBA.

Microphone Handling Noise: Measurements of Perceptual Threshold and Effects on Audio Quality.

A psychoacoustic experiment was carried out to test the effects of microphone handling noise on perceived audio quality. Handling noise is a problem affecting both amateurs using their smartphones and cameras, as well as professionals using separate microphones and digital recorders. The noises used for the tests were measured from a variety of devices, including smartphones, laptops and handheld microphones. The signal features that characterise these noises are analysed and presented. The sounds include various types of transient, impact noises created by tapping or knocking devices, as well as more sustained sounds caused by rubbing. During the perceptual tests, listeners auditioned speech podcasts and were asked to rate the degradation of any unwanted sounds they heard. A representative design test methodology was developed that tried to encourage everyday rather than analytical listening. Signal-to-noise ratio (SNR) of the handling noise events was shown to be the best predictor of quality degradation. Other factors such as noise type or background noise in the listening environment did not significantly affect quality ratings. Podcast, microphone type and reproduction equipment were found to be significant but only to a small extent. A model allowing the prediction of degradation from the SNR is presented. The SNR threshold at which 50% of subjects noticed handling noise was found to be 4.2 ± 0.6 dBA. The results from this work are important for the understanding of our perception of impact sound and resonant noises in recordings, and will inform the future development of an automated predictor of quality for handling noise.

Fast time domain modeling of surface scattering from reflectors and diffusers.

Time-domain prediction models have been developed for auditorium reflectors and room acoustic diffusers. The models are time-domain equivalents of the single-frequency formulations that exploit the Kirchhoff boundary conditions. Consequently, they are approximate, wave-based solutions to the Kirchhoff integral equation using surface meshes. The new time-domain formulations are validated by comparison to their frequency-domain equivalents for three different surfaces: a plane surface, a curved reflector, and a Schroeder diffuser. In terms of computation time and accuracy, the new models lie between the finite difference time domain and geometric room models.

A record "longest echo" within the Inchindown oil despository.

In 2013, Guinness World Records awarded tank number 1 at the Inchindown oil despository, Ross-shire, Scotland, the record for the "longest echo" at 75 s. Guinness World Records calls it the longest echo because that was the name of the record that was broken, however, the correct name for the phenomenon measured is reverberation. This Letter has been written to document this unique acoustic space where the reverberation time is 112 s at 125 Hz, to detail the methodology for those who wish to attempt to break the record, and to discuss why the tank is so uniquely reverberant.

Perception and automatic detection of wind-induced microphone noise.

Wind can induce noise on microphones, causing problems for users of hearing aids and for those making recordings outdoors. Perceptual tests in the laboratory and via the Internet were carried out to understand what features of wind noise are important to the perceived audio quality of speech recordings. The average A-weighted sound pressure level of the wind noise was found to dominate the perceived degradation of quality, while gustiness was mostly unimportant. Large degradations in quality were observed when the signal to noise ratio was lower than about 15 dB. A model to allow an estimation of wind noise level was developed using an ensemble of decision trees. The model was designed to work with a single microphone in the presence of a variety of foreground sounds. The model outputted four classes of wind noise: none, low, medium, and high. Wind free examples were accurately identified in 79% of cases. For the three classes with noise present, on average 93% of samples were correctly assigned. A second ensemble of decision trees was used to estimate the signal to noise ratio and thereby infer the perceived degradation caused by wind noise.

Adolescents' perceptions of their school's acoustic environment: the development of an evidence based questionnaire.

A poor acoustic environment in a school is known to negatively affect pupils' learning and achievement. This paper presents the design and findings of an online questionnaire survey of 11-16 year olds' impressions of their school's acoustic environment. A total of 2588 English secondary school pupils responded to the questionnaire. Factor analysis was used to identify variables which best characterized pupils' impressions of their school's acoustic environment. Four factors, corresponding to ease of hearing in school spaces, sensitivity to noise, the consequences of noise in the classroom, and annoyance to intermittent noise, accounted for 43% of the total variance in pupils' responses to the questionnaire. Analysis of the responses on these factors showed that pupils who reported additional learning needs such as hearing impairment, speaking English as an additional language or receiving learning support reported being significantly more affected by poor school acoustics than pupils reporting no additional learning needs. Older pupils were significantly more sensitive to noise annoyance and to the consequences of poor acoustical conditions on their learning and behaviour than younger pupils. Pupils attending suburban schools featuring cellular classrooms that were not exposed to a nearby noise sources were more positive about their school acoustics than pupils at schools with open plan classroom designs or attending schools that were exposed to external noise sources. The study demonstrates that adolescents are reliable judges of their school's acoustic environment, and have insight into the disruption to teaching and learning caused by poor listening conditions. Furthermore, pupils with additional learning needs are more at risk from the negative effects of poor school acoustics.

Acoustic iridescence.

An investigation has been undertaken into acoustic iridescence, exploring how a device can be constructed which alter sound waves, in a similar way to structures in nature that act on light to produce optical iridescence. The main construction had many thin perforated sheets spaced half a wavelength apart for a specified design frequency. The sheets create the necessary impedance discontinuities to create backscattered waves, which then interfere to create strongly reflected sound at certain frequencies. Predictions and measurements show a set of harmonics, evenly spaced in frequency, for which sound is reflected strongly. And the frequency of these harmonics increases as the angle of observation gets larger, mimicking the iridescence seen in natural optical systems. Similar to optical systems, the reflections become weaker for oblique angles of reflection. A second construction was briefly examined which exploited a metamaterial made from elements and inclusions which were much smaller than the wavelength. Boundary element method predictions confirmed the potential for creating acoustic iridescence from layers of such a material.

Volumetric diffusers: pseudorandom cylinder arrays on a periodic lattice.

Most conventional diffusers take the form of a surface based treatment, and as a result can only operate in hemispherical space. Placing a diffuser in the volume of a room might provide greater efficiency by allowing scattering into the whole space. A periodic cylinder array (or sonic crystal) produces periodicity lobes and uneven scattering. Introducing defects into an array, by removing or varying the size of some of the cylinders, can enhance their diffusing abilities. This paper applies number theoretic concepts to create cylinder arrays that have more even scattering. Predictions using a boundary element method are compared to measurements to verify the model, and suitable metrics are adopted to evaluate performance. Arrangements with good aperiodic autocorrelation properties tend to produce the best results. At low frequency power is controlled by object size and at high frequency diffusion is dominated by lattice spacing and structural similarity. Consequently the operational bandwidth is rather small. By using sparse arrays and varying cylinder sizes, a wider bandwidth can be achieved.

A transient boundary element method model of Schroeder diffuser scattering using well mouth impedance.

Room acoustic diffusers can be used to treat critical listening environments to improve sound quality. One popular class is Schroeder diffusers, which comprise wells of varying depth separated by thin fins. This paper concerns a new approach to enable the modeling of these complex surfaces in the time domain. Mostly, diffuser scattering is predicted using steady-state single frequency methods. A popular approach is to use a frequency domain boundary element method (BEM) model of a box containing the diffuser, where the mouth of each well is replaced by a compliant surface with appropriate surface impedance. The best way of representing compliant surfaces in time domain prediction models, such as the transient BEM is, however, currently unresolved. A representation based on surface impedance yields convolution kernels which involve future sound, so is not compatible with the current generation of time-marching transient BEM solvers. Consequently, this paper proposes the use of a surface reflection kernel for modeling well behavior and this is tested in a time domain BEM implementation. The new algorithm is verified on two surfaces including a Schroeder diffuser model and accurate results are obtained. It is hoped that this representation may be extended to arbitrary compliant locally reacting materials.

Monaural room acoustic parameters from music and speech.

This paper compares two methods for extracting room acoustic parameters from reverberated speech and music. An approach which uses statistical machine learning, previously developed for speech, is extended to work with music. For speech, reverberation time estimations are within a perceptual difference limen of the true value. For music, virtually all early decay time estimations are within a difference limen of the true value. The estimation accuracy is not good enough in other cases due to differences between the simulated data set used to develop the empirical model and real rooms. The second method carries out a maximum likelihood estimation on decay phases at the end of notes or speech utterances. This paper extends the method to estimate parameters relating to the balance of early and late energies in the impulse response. For reverberation time and speech, the method provides estimations which are within the perceptual difference limen of the true value. For other parameters such as clarity, the estimations are not sufficiently accurate due to the natural reverberance of the excitation signals. Speech is a better test signal than music because of the greater periods of silence in the signal, although music is needed for low frequency measurement.

Lüke and power residue sequence diffusers.

Conventional Schroeder diffusers have been successfully used for many years. However, their frequency range is limited by the flat plate effect that occurs when all the wells radiate in phase. This occurs at harmonics of p times the design frequency f(0), where p is the small prime that is used to generate the structure. A typical diffuser, using p=7 and f(0)=500 Hz, has an upper frequency limit of only 3.5 kHz. Achieving a first flat plate frequency above 20 kHz requires a prime equal to at least 41 and results in diffusers that are too big to be practical in most applications. This paper suggests an alternative approach using number theoretic sequences that, although short in length, are based on large integers. Two new sequences, Type-II Luke and power residue, have this desired characteristic. They are investigated using both simple models and the more exact boundary element method. The results show the flat plate effect is moved to much higher frequencies as expected. For Luke sequences at certain frequencies, redirection rather than dispersion is achieved. Modulation techniques can be used to mitigate these problems. Power residue sequences perform the best, providing good diffusion and a flat plate frequency outside the audible range.

Event-related potential study to aversive auditory stimuli.

In an auditory oddball task emotionally negative (aversive) sounds (e.g. rubbing together of polystyrene) and everyday sounds (e.g. ringing of a bicycle bell) were presented as task-irrelevant (novel) sounds. Both the aversive and the everyday sounds elicited the orientation-related P3a component of the event-related potentials (ERPs). In the 154-250 ms range the ERPs for the aversive sounds were more negative than the ERP of the everyday sounds. For the aversive sounds, this negativity was followed by a frontal positive wave (372-456 ms). The aversive sounds elicited larger late positive shift than the everyday sounds. The early negativity is considered as an initial effect in a broad neural network including limbic structures, while the later is related to the cognitive assessment of the stimuli and to memory-related processes.

Ternary and quadriphase sequence diffusers.

A room acoustic diffuser breaks up reflected wavefronts, and this can be achieved by presenting a spatially varying surface impedance. In hybrid surfaces, varying impedance is achieved by patches of absorption and reflection, giving reflection coefficients nominally of 0 and 1. These surfaces are hybrids, absorbing some of the incident sound while diffusing any reflected energy. A problem with planar hybrid surfaces is that specular energy is only removed by absorption. By exploiting interference, by reflecting waves out-of-phase with the specular energy, it is possible to diminish the specular energy further. This can be achieved by using a diffuser based on a ternary sequence that nominally has reflection coefficients of 0, -1, and +1. Ternary sequences are therefore a way of forming hybrid absorber-diffusers that achieve better scattering performance without additional absorption. This paper discusses methods for making ternary sequence diffusers, including giving sequence generation methods. It presents prediction results based on Fourier and boundary element method models to examine the performance. While ternary diffusers have better performance than unipolar binary diffusers at most frequencies, there are frequencies at which the performances are the same. This can be overcome by forming diffusers from four-level, quadriphase sequences.