Subjective Listening Tests Research Articles

Subjective Audio Quality Testing, With Tasting and Car Driving as Parallel Tasks

This paper reports subjective test results of two different test sets: one without and one with a parallel task. The subjective listening tests were performed according to ITU-P.800 Recommendation. Regarding the parallel tasks, there were two categories: a parallel task that requires visual/motor coordination; and another that requires the use of taste perception. Specifically, in the first experiment, subjects were asked to drive a car simulator, while in the second they were asked to sort a variety of samples by taste. In both situations, the subjects performed the parallel task while assessing the audio quality. The results obtained in the experiments with a parallel task showed considerable differences from the standard P.800 test. Nevertheless, similar features were observed in the results of both parallel task tests. Based on those results, an objective estimator of speech quality assessed with a parallel task was developed, which was based on an enhanced ITU-T P.863 (POLQA) algorithm. The improved estimator provides a more accurate prediction of subjective tests with a parallel task than the original POLQA algorithm, as shown also in the results.

Direct Speech Reconstruction From Articulatory Sensor Data by Machine Learning

This paper describes a technique that generates speech acoustics from articulator movements. Our motivation is to help people who can no longer speak following laryngectomy, a procedure that is carried out tens of thousands of times per year in the Western world. Our method for sensing articulator movement, permanent magnetic articulography, relies on small, unobtrusive magnets attached to the lips and tongue. Changes in magnetic field caused by magnet movements are sensed and form the input to a process that is trained to estimate speech acoustics. In the experiments reported here this “Direct Synthesis” technique is developed for normal speakers, with glued-on magnets, allowing us to train with parallel sensor and acoustic data. We describe three machine learning techniques for this task, based on Gaussian mixture models, deep neural networks, and recurrent neural networks (RNNs). We evaluate our techniques with objective acoustic distortion measures and subjective listening tests over spoken sentences read from novels (the CMU Arctic corpus). Our results show that the best performing technique is a bidirectional RNN (BiRNN), which employs both past and future contexts to predict the acoustics from the sensor data. BiRNNs are not suitable for synthesis in real time but fixed-lag RNNs give similar results and, because they only look a little way into the future, overcome this problem. Listening tests show that the speech produced by this method has a natural quality that preserves the identity of the speaker. Furthermore, we obtain up to 92% intelligibility on the challenging CMU Arctic material. To our knowledge, these are the best results obtained for a silent-speech system without a restricted vocabulary and with an unobtrusive device that delivers audio in close to real time. This work promises to lead to a technology that truly will give people whose larynx has been removed their voices back.

Restricted Boltzmann Machine-Based Voice Conversion for Nonparallel Corpus

A large amount of parallel training corpus is necessary for robust, high-quality voice conversion. However, such parallel data may not always be available. This letter presents a new voice conversion method that needs no parallel speech corpus, and adopts a restricted Boltzmann machine (RBM) to represent the distribution of the spectral features derived from a target speaker. A linear transformation was employed to convert the spectral and delta features. A conversion function was obtained by maximizing the conditional probability density function with respect to the target RBM. A feasibility test was carried out on the OGI VOICES corpus. Results from the subjective listening tests and the objective results both showed that the proposed method outperforms the conventional GMM-based method.

New horizons of vehicle aerodynamics

Vehicle aerodynamics and wind tunnel technology are progressing towards more realistic simulations of the real-world on-road environment. This paper presents an overview of the new systems which were implemented during the recent wind tunnel upgrade at Forschungsinstitut für Kraftfahrwesenund Fahrzeugmotoren Stuttgart as well as comparable computational fluid dynamics simulations. The fully interchangeable road simulation system features an interchangeable five-belt system and three-belt system in the same full-scale automotive wind tunnel. This system offers the efficiency of a five-belt system combined with the more sophisticated ground simulation technique of a wide belt system, which is necessary to assess the aerodynamic properties of sports cars and racing cars. In order to simulate on-road wind conditions, a side-wind generator can be installed to generate a turbulent flow field in the wind tunnel test section. It could be shown that the commonly determined drag coefficient at 0° yaw angle in the smooth flow environment of today’s wind tunnels is not representative of the drag found in real on-road wind conditions. Additionally, the investigations in unsteady side-wind conditions indicate that the commonly used approach to determine the side-wind sensitivity of a vehicle underestimates the forces occurring in turbulent flow conditions. A validated simulation model is presented. The simulation results are in good agreement with the experimental results and can be used as a complementary tool when assessing the unsteady aerodynamic behaviour of a vehicle; this behaviour can be coupled to a vehicle dynamics model for virtual road testing in the Stuttgart full-motion driving simulator. The unsteady-behaviour effects can be evaluated comprehensively, and the results allow a subjective assessment of the unsteady response of the vehicle. Furthermore, the aeroacoustic wind noise in on-road wind conditions is investigated during the development of the vehicle. The side-wind generator reproduces the natural stochastic cross-wind and allows the effect of these wind conditions to be investigated in the aeroacoustic wind tunnel. The results show similar ratings to those in on-road tests when compared with subjective listening tests. In summary, the techniques introduced open up new horizons in the field of vehicle aerodynamics and aeroacoustics, which are a step closer towards real-world conditions in automotive engineering.

Audio enhancement using local SNR-based sparse binary mask estimation and spectral imputation

This paper proposes a method for enhancing speech and/or audio quality under noisy conditions. The proposed method first estimates the local signal-to-noise ratio (SNR) of the noisy input signal via sparse non-negative matrix factorization (SNMF). Next, a sparse binary mask (SBM) is proposed that separates the audio signal from the noise by measuring the sparsity of the pool of local SNRs from the adjacent frequency bands of the current and several previous frames. However, some spectral gaps remain across frequency bands after applying the binary masks, which distorts the separated audio signal due to spectral discontinuity. Thus, a spectral imputation technique is used to fill the empty spectrum of the frequency band where it is removed by the SBM. Spectral imputation is conducted by online learning NMF with the spectra of the neighboring non-overlapped frequency bands and their local sparsity. The effectiveness of the proposed enhancement method is demonstrated on two different tasks use speech and musical content, respectively. Consequently, objective measurements and subjective listening tests show that the proposed method outperforms conventional speech and audio enhancement methods, such as SNMF-based alternatives and deep recurrent neural networks for speech enhancement, block thresholding, and a commercially available software tool for audio enhancement.

Sound image externalization for headphone based real-time 3D audio

3D audio effects can provide immersive auditory experience, but we often face the so-called in-head localization (IHL) problem in headphone sound reproduction. To address this problem, we propose an effective sound image externalization approach. Specifically, we consider several important factors related to sound propagation, which include image-source model based early reflections with distance decay, wall absorption and air absorption, late reverberation and other dynamic factors like head movement. We apply our sound image externalization approach to a headphone based real-time 3D audio system. Subjective listening tests show that the sound image externalization performance is significantly improved and the sound source direction is preserved as well. A/B preference test further shows that, as compared with a recent popular approach, the proposed approach is mostly preferred by the listeners.

A Novel Mobility-Aware Playout Algorithm for VoIP Services

The latest explosive growth in mobile networks has resulted in an increasing interest in optimisation techniques for mobile services. Resent advances in mobile wireless networks incorporate link-layer intelligence in order to enhance the performance of network and application layers. The Media Independent Handover (MIH) standard provides a framework that can make such link-layer intelligence available to upper layers. In this paper, a novel MIH-enabled playout algorithm for Voice over Internet Protocol applications is presented that aims to compensate for the degradation of voice quality caused by handovers. To that end, link-layer triggers in conjunction with speech time-scale modification techniques are exploited to mitigate the increase in delay and jitter induced by the handover process. Results of subjective listening tests show typical gains of 0.3 on a 5-point scale of the Mean Opinion Score with respect to existing playout scheduling schemes.

Multichannel speech reinforcement based on binaural unmasking

Speech reinforcement or near-end listening enhancement is a technique that modifies the far-end signal to mitigate the effect of the near-end noise, usually based on the power spectra of the far-end signal and the near-end noise. Psychoacoustic experiments have shown that the location of a noise source with respect to that of a signal source affects the amount of masking. Since conventional speech reinforcement methods obtain spectral gain based only on the power spectra, this psychoacoustic phenomenon called binaural unmasking has not been considered in those approaches. In this paper, we propose a novel speech reinforcement algorithm that modifies the far-end speech signal based on both the power spectrum and the direction-of-arrival (DoA) of the noise. Specifically, we have computed the equivalent frontal noise level from the observed noise level and the estimated DoA, and used it to compute spectral gains as in conventional partial loudness restoration-based speech reinforcement. Experimental results showed that the proposed method outperformed the conventional methods based on partial loudness restoration and speech intelligibility index (SII) optimization in terms of the overall perceived quality through subjective listening tests.

Subjective listening tests: Perception and preference of simulated sound fields

Concert hall sound fields were simulated by architecture students and anechoic recordings were convolved to create auralizations in those simulated performance spaces. Then an architectural feature was altered digitally and a second track was auralized. College music students were recruited, tested for hearing loss, and brought to a low-reverberance room with a spatial sound array of 28 mounted speakers. They were asked to identify which of the two simulated tracks they prefer. We compared simulated performance spaces: (1) with four tiers of balconies vs with one tier of balcony; (2) with an over-stage canopy vs without a canopy; (3) with separate balcony boxes vs with a continuous balcony not fragmented by box walls; and (4) with a higher scattering coefficient vs a lower scattering coefficient. Those in the audience will be invited to judge preference between the tracks for themselves. The study will be framed by the extraordinary career arc of Bert Kinzey who engaged architecture students in the study of architectural acoustics at both Virginia Tech and at the University of Florida.

The Reduction of Vertical Interchannel Crosstalk: The Analysis of Localisation Thresholds for Natural Sound Sources

In subjective listening tests, natural sound sources were presented to subjects as vertically-oriented phantom images from two layers of loudspeakers, ‘height’ and ‘main’. Subjects were required to reduce the amplitude of the height layer until the position of the resultant sound source matched that of the same source presented from the main layer only (the localisation threshold). Delays of 0, 1 and 10 ms were applied to the height layer with respect to the main, with vertical stereophonic and quadraphonic conditions being tested. The results of the study showed that the localisation thresholds obtained were not significantly affected by sound source or presentation method. Instead, the only variable whose effect was significant was interchannel time difference (ICTD). For ICTD of 0 ms, the median threshold was −9.5 dB, which was significantly lower than the −7 dB found for both 1 and 10 ms. The results of the study have implications both for the recording of sound sources for three-dimensional (3D) audio reproduction formats and also for the rendering of 3D images.

Comparison of single-microphone noise reduction schemes: can hearing impaired listeners tell the difference?

Objective: The perceived qualities of nine different single-microphone noise reduction (SMNR) algorithms were to be evaluated and compared in subjective listening tests with normal hearing and hearing impaired (HI) listeners. Design: Speech samples added with traffic noise or with party noise were processed by the SMNR algorithms. Subjects rated the amount of speech distortions, intrusiveness of background noise, listening effort and overall quality, using a simplified MUSHRA (ITU-R, 2003) assessment method. Study sample: 18 normal hearing and 18 moderately HI subjects participated in the study. Results: Significant differences between the rating behaviours of the two subject groups were observed: While normal hearing subjects clearly differentiated between different SMNR algorithms, HI subjects rated all processed signals very similarly. Moreover, HI subjects rated speech distortions of the unprocessed, noisier signals as being more severe than the distortions of the processed signals, in contrast to normal hearing subjects. Conclusions: It seems harder for HI listeners to distinguish between additive noise and speech distortions or/and they might have a different understanding of the term “speech distortion” than normal hearing listeners have. The findings confirm that the evaluation of SMNR schemes for hearing aids should always involve HI listeners.

Validation of a Numerical Model for the Prediction of the Annoyance Condition at the Operator Station of Construction Machines

It is well-known that the reduction of noise levels is not strictly linked to the reduction of noise annoyance. Even earthmoving machine manufacturers are facing the problem of customer complaints concerning the noise quality of their machines with increasing frequency. Unfortunately, all the studies geared to the understanding of the relationship between multidimensional characteristics of noise signals and the auditory perception of annoyance require repeated sessions of jury listening tests, which are time-consuming. In this respect, an annoyance prediction model was developed for compact loaders to assess the annoyance sensation perceived by operators at their workplaces without repeating the full sound quality assessment but using objective parameters only. This paper aims at verifying the feasibility of the developed annoyance prediction model when applied to other kinds of earthmoving machines. For this purpose, an experimental investigation was performed on five earthmoving machines, different in type, dimension, and engine mechanical power, and the annoyance predicted by the numerical model was compared to the annoyance given by subjective listening tests. The results were evaluated by means of the squared value of the correlation coefficient, R2, and they confirm the possible applicability of the model to other kinds of machines.

A study on the effect by adding reverberation to speech-like masking sound

Problems arising from the acoustical privacy point of view in public spaces have been known to be an issue. The lack of acoustical privacy has been known to affect the human's health both physically and psychologically, thus keeping the acoustical privacy in public spaces will significantly reduce social loss. Masking is the most commonly practically used technique to make a target speech unintelligible to the unintended listeners without needing to install any physical structures. Time-reversed speech has been known to effectively mask information for speech privacy applications; however, the annoyance and distraction caused by the time-reversed speech is known to be higher than other masking sound. This study explores a solution to compromise the suggested problem by adding a reverberant effect to a time-reversed speech. Subjective listening tests have been conducted to measure the intelligibility of target speech, annoyance and distraction caused by the masking sound. The experimental results suggest that adding artificial reverberation to a time-reversed speech has a significant effect to reduce the annoyance level while maintaining the masking effectiveness of the original time-reversed speech. A trend was also observed that the addition of artificial reverberation could reduce the level of distraction caused by the masking sound.

Sound source separation and synthesis for audio enhancement based on spectral amplitudes of two-channel stereo signals

A sound source separation algorithm based on the spectral amplitudes of 2-channel signals has been developed for the up-mixing playback of 2-channel stereo. Short-term Fourier transforms (STFT) of the signals on the left and right channels are first calculated. The coefficients of the discrete Fourier transform (DFT) are used to calculate the ratio of the spectral amplitudes of the left and right channels, which is termed the channel level difference (CLD). The DFT coefficients are then divided into multiple groups on the basis of the CLD, with each group representing a separated sound source. The signal-to-distortion ratio (SDR)is used to evaluate the signal separation performance. It was found that a rough estimate of the CLD threshold yielding the best SDR could be obtained by cross-correlating the separated sounds. For playback on a headset, each separated signal is convoluted with head-related transfer functions (HRTF) that represent the direction of that particular sound source. Subjective listening tests showed that the sound synthesized by this method is more realistic than that synthesized with HRTFs that represent only left and right speakers.

Perceptually plausible acoustics simulation of single and coupled rooms

The fast and perceptually plausible room acoustics simulator [RAZR, see Wendt et al., JAES 62, 11 (2014)] has been optimized and extended to simulate the acoustics of connected rooms. RAZR synthesizes binaural room impulse responses (BRIRs) with high computational efficiency using a hybrid approach: early reflections are calculated as image sources for a shoebox-room approximation up to a low order, and the later reverberation is generated by a binaurally extended feedback delay network (FDN). For the extension toward two coupled rooms, validity, visibility, and diffraction of the image sources in the two rooms were taken into account, as well as different reverberation properties and the opening angle of the door to the neighbor room seen from a specific receiver position. The suggested method was evaluated by comparing measured and synthesized BRIRs for several rooms differing in size and reverberation time. For room acoustical parameters such as early decay time, late decay time, definition, and clarity a good agreement in terms of the Pearson correlation coefficient (0.86—0.99) was achieved between measured and simulated BRIRs. Subjective listening tests showed a good agreement of ratings of perceptual attributes (e.g., tone-color, reverberation time, envelopment, and naturalness) for the measured and synthesized BRIRs.

Subjective Evaluation of Three Headphone-Based Virtual Sound Source Positioning Methods Including Differential Head-Related Transfer Function

Abstract This paper analyses the performance of Differential Head-Related Transfer Function (DHRTF), an alternative transfer function for headphone-based virtual sound source positioning within a horizontal plane. This experimental one-channel function is used to reduce processing and avoid timbre affection while preserving signal features important for sound localisation. The use of positioning algorithm employing the DHRTF is compared to two other common positioning methods: amplitude panning and HRTF processing. Results of theoretical comparison and quality assessment of the methods by subjective listening tests are presented. The tests focus on distinctive aspects of the positioning methods: spatial impression, timbre affection, and loudness fluctuations. The results show that the DHRTF positioning method is applicable with very promising performance; it avoids perceptible channel coloration that occurs within the HRTF method, and it delivers spatial impression more successfully than the simple amplitude panning method.

Effect of adding artificial reverberation to speech-like masking sound

Time-reversed speech has been known to effectively mask information for speech privacy applications. However, the annoyance and distraction caused by the time-reversed speech-like masking sound is higher than other masking sound. This study investigates the effects of adding artificial reverberation to the time-reversed speech. Subjective listening tests have been conducted to measure the intelligibility of target speech, annoyance and distraction caused by the masking sound. The experimental results suggest that adding artificial reverberation to a speech-like masking sound has a significant effect to reduce the annoyance level while maintaining the masking effectiveness of the original masking sound. A trend was also observed that the addition of artificial reverberation could reduce the level of distraction caused by the masking sound.

Audio bandwidth extension using ensemble of recurrent neural networks

In audio communication systems, the perceptual audio quality of the reproduced audio signals such as the naturalness of the sound is limited by the available audio bandwidth. In this paper, a wideband to super-wideband audio bandwidth extension method is proposed using an ensemble of recurrent neural networks. The feature space of wideband audio is firstly divided into different regions through clustering. For each region in the feature space, a specific recurrent neural network with a sparsely connected hidden layer, referred as the echo state network, is employed to dynamically model the mapping relationship between wideband audio features and high-frequency spectral envelope. In the following step, the outputs of multiple echo state networks are weighted and fused by means of network ensemble, in order to further estimate the high-frequency spectral envelope. Finally, combining the high-frequency fine spectrum extended by spectral translation, the proposed method can effectively extend the bandwidth of wideband audio to super wideband. Objective evaluation results show that the proposed method outperforms the hidden Markov model-based bandwidth extension method on the average in terms of both static and dynamic distortions. In subjective listening tests, the results indicate that the proposed method is able to improve the auditory quality of the wideband audio signals and outperforms the reference method.

SPEECH ENHANCEMENT USING SPECTRAL SUBTRACTION TECHNIQUE WITH MINIMIZED CROSS SPECTRAL COMPONENTS

The aim of speech enhancement is to get significant reduction of noise and enhanced speech from noisy speech. There are several approaches for speech enhancement .earlier approaches didn’t consider cross spectral terms into account. Cross spectral terms become prominent when processing window size becomes small i.e. 20ms-30ms. In this paper, an enhancement method is proposed for significant reduction of noise, and improvement in the quality and perceptibility of speech degraded by correlated additive background noise. The proposed method is based on the spectral subtraction technique. The simple spectral subtraction technique results in poor reduction of noise. One of the main reasons for this is neglecting the cross spectral terms of speech and noise, based on the appropriation that clean speech and noise signals are completely uncorrelated to each other, which is not true on short time basis. In this paper an improvement in reduction of the noise is achieved as compared to the earlier methods. This fact is mainly attributed to the cross spectral terms between speech and noise. This algorithm can be implemented and used in hearing aids for the benefit of hearing impaired people. Objective speech quality measures, spectrogram analyses and subjective listening tests conforms the proposed method is more effective in comparison with earlier speech enhancement techniques.

Audio Bandwidth Extension Based on Ensemble Echo State Networks with Temporal Evolution

The bandwidth limitation of wideband audio systems degrades the subjective quality and naturalness of audio signals. In this paper, a new method for blind bandwidth extension of wideband audio signals is proposed based on ensemble echo state network with temporal evolution. The high-frequency components in the band of $7 \sim 14 \text{kHz}$ are artificially restored only from the information in the wideband audio. For each region in the wideband feature space, a specific echo state network with recurrent structure is explored to dynamically model the local mapping relationship between wideband audio features and high-frequency spectral envelope. The transition process among regions is modeled by a hidden Markov model, and a network ensemble technique based on temporal evolution is used to fuse multiple echo state networks such that the high-frequency spectral envelope is estimated. Combining the high-frequency fine spectrum extended by spectral translation, the proposed method can effectively extend the wideband audio to super wideband. In addition, the proposed extension method is applied to the ITU-T G.729.1 wideband audio codec and is further evaluated in comparison with the ITU-T G.729.1 Annex E super-wideband audio codec and the hidden Markov model-based reference bandwidth extension method. Objective quality evaluation results indicate that the proposed method is preferred over the hidden Markov model-based reference bandwidth extension method in terms of log spectral distortion, cosh measure, and differential log spectral distortion. Further, the proposed method improves the auditory quality of the wideband audio and also gains a good performance in the subjective listening tests.