Vocal sound imitations provide a new challenge for understanding the coupling between articulatory mechanisms and the resulting audio. In this study, we have modeled the classification of three articulatory categories, phonation, supraglottal myoelastic vibrations, and turbulence from audio recordings. Two data sets were assembled, consisting of different vocal imitations by four professional imitators and four non-professional speakers in two different experiments. The audio data were manually annotated by two experienced phoneticians using a detailed articulatory description scheme. A separate set of audio features was developed specifically for each category using both time-domain and spectral methods. For all time-frequency transformations, and for some secondary processing, the recently developed Auditory Receptive Fields Toolbox was used. Three different machine learning methods were applied for predicting the final articulatory categories. The result with the best generalization was found using an ensemble of multilayer perceptrons. The cross-validated classification accuracy was 96.8 % for phonation, 90.8 % for supraglottal myoelastic vibrations, and 89.0 % for turbulence using all the 84 developed features. A final feature reduction to 22 features yielded similar results.

A unified approach for the numerical simulation of vowels is presented, which accounts for the self-oscillations of the vocal folds including contact, the generation of acoustic waves and their propagation through the vocal tract, and the sound emission outwards the mouth. A monolithic incompressible fluid-structure interaction model is used to simulate the interaction between the glottal jet and the vocal folds, whereas the contact model is addressed by means of a level set application of the Eikonal equation. The coupling with acoustics is done through an acoustic analogy stemming from a simplification of the acoustic perturbation equations. This coupling is one-way in the sense that there is no feedback from the acoustics to the flow and mechanical fields. All the involved equations are solved together at each time step and in a single computational run, using the finite element method (FEM). As an application, the production of vowel [i] has been addressed. Despite the complexity of all physical phenomena to be simulated simultaneously, which requires resorting to massively parallel computing, the formant locations of vowel [i] have been well recovered.

Although it has been shown in previous research (Orlikoff, 1991; Henrich et al, 2005; Kuang et al, 2014; Awan, 2015) that there exists a relationship between the electroglottogram (EGG) waveform and the acoustic signal, this relationship is still not fully understood. To investigate this relationship, the EGG and acoustic signals were measured for four male amateur choir singers who each produced eight consecutive tones of increasing and decreasing vocal intensity. The EGG signals were processed cycle-synchronously to obtain the discrete Fourier transform, and the data were used as an input to a clustering algorithm. The acoustic signal was analyzed in terms of sound pressure level (dB SPL) and fundamental frequency (f(o)) of vibration, and the results of both EGG and acoustic analysis were depicted on a two-dimensional plane with f(o) on the x-axis and SPL on the y-axis. All the subjects were seen to have a weak, near-sinusoidal EGG waveform in their lowest SPL range, whereas increase in SPL coincided with progressive enrichment in harmonic content of the EGG waveforms. The results of the clustering were additionally used to classify waveforms across subjects to enable inter-subject comparisons and assessment of individual strategies of exploring the f(o)-SPL dimensions. In these male subjects, the EGG waveform shape appeared to vary with SPL and to remain essentially constant with f(o) over one octave.

Objectives

Good voice quality is an asset to professional voice users, including radio performers. We examined whether (1) voices could be reliably categorized as good for the radio and (2) these categories could be predicted using acoustic measures.

Participants and Methods

Male radio performers (n = 24) and age-matched male controls performed “The Rainbow Passage” as if presenting on the radio. Voice samples were rated using a three-stage paired-comparison paradigm by 51 naive listeners and perceptual categories were identified (Study 1), and then analyzed for fundamental frequency, long-term average spectrum, cepstral peak prominence, and pause or spoken-phrase duration (Study 2).

Results

Study 1: Good inter-judge reliability was found for perceptual judgments of the best 15 voices (good for radio category, 14/15 = radio performers), but agreement on the remaining 33 voices (unranked category) was poor. Study 2: Discriminant function analyses showed that the SD standard deviation of sounded portion duration, equivalent sound level, and smoothed cepstral peak prominence predicted membership of categories with moderate accuracy (R² = 0.328).

Conclusions

Radio performers are heterogeneous for voice quality; good voice quality was judged reliably in only 14 out of 24 radio performers. Current acoustic analyses detected some of the relevant signal properties that were salient in these judgments. More refined perceptual analysis and the use of other perceptual methods might provide more information on the complex nature of judging good voices.

Objectives: To investigate how the direct biofeedback on vocal loudness administered with a portable voice accumulator (VoxLog) should be configured, to facilitate an optimal learning outcome for individuals with Parkinson's disease (PD), on the basis of principles of motor learning. Study Design: Methodologic development in an experimental study. Methods: The portable voice accumulator VoxLog was worn by 20 participants with PD during habitual speech during semistructured conversations. Six different biofeedback configurations were used, in random order, to study which configuration resulted in a feedback frequency closest to 20% as recommended on the basis of previous studies. Results: Activation of feedback when the wearer speaks below a threshold level of 3dB below the speaker's mean voice sound level in habitual speech combined with an activation time of 500ms resulted in a mean feedback frequency of 21.2%. Conclusions: Settings regarding threshold and activation time based on the results from this study are recommended to achieve an optimal learning outcome when administering biofeedback on vocal loudness for individuals with PD using portable voice accumulators.

An overview is given of the current status of the computerised voice range profile (VRP) as a voice measurement paradigm. Its operating principles are described, and sources of errors and variability are discussed. The features of the VRP contour and its characterisaï¿œtion are described. Methods for performing statistics on VRP contour and interior data are considered. Examples are given of clinical, pedagogical and research applications. Finally, issues with the models used to interpret VRP data are discussed. It is concluded that, while the VRP offers a convenient frame of reference for a multitude of voice assessment metrics, it also exposes the many degrees of freedom in the voice to an extent that challenges us to improve our models of how the voice functions over a large range and in a dynamic setting.

Objectives and Study Design. Information about how patients with voice disorders use their voices in natural communicative situations is scarce. Such long-term data have for the first time been uploaded to a central database from different hospitals in Sweden. The purpose was to investigate the potential use of a large set of long-term data for establishing reference values regarding voice use in natural situations. Methods. VoxLog (Sonvox AB, Umeå, Sweden) was tested for deployment in clinical practice by speech-language pathologists working at nine hospitals in Sweden. Files from 20 patients (16 females and 4 males) with functional, organic, or neurological voice disorders and 10 vocally healthy individuals (eight females and two males) were uploaded to a remote central database. All participants had vocally demanding occupations and had been monitored for more than 2 days. The total recording time was 681 hours and 50 minutes. Data on fundamental frequency (F0, Hz), phonation time (seconds and percentage), voice sound pressure level (SPL, dB), and background noise level (dB) were analyzed for each recorded day and compared between the 2 days. Variations across each day were measured using coefficients of variation. Results. Average F0, voice SPL, and especially the level of background noise varied considerably for all participants across each day. Average F0 and voice SPL were considerably higher than reference values from laboratory recordings. Conclusions. The use of a remote central database and strict protocols can accelerate data collection from larger groups of participants and contribute to establishing reference values regarding voice use in natural situations and from patients with voice disorders. Information about activities and voice symptoms would supplement the objective data and is recommended in future studies.

Voice and gestures are natural sketching tools that can be exploited to communicate sonic interactions. In product and interaction design, sounds should be included in the early stages of the design process. Scientists of human motion have shown that auditory stimuli are important in the performance of difficult tasks and can elicit anticipatory postural adjustments in athletes. These findings justify the attention given to sound in interaction design for gaming, especially in action and sports games that afford the development of levels of virtuosity. The sonic manifestations of objects can be designed by acting on their mechanical qualities and by augmenting the objects with synthetic and responsive sounds.

The vocal folds can oscillate in several different ways, manifest to practitioners and clinicians as ‘registers’ or ‘mechanisms’, of which the two most commonly considered are modal voice and falsetto voice. Here these will be taken as instances of different ‘vibratory states’, i.e., distinct quasi-stationary patterns of vibration of the vocal folds. State transitions are common in biomechanical nonlinear oscillators; and they are often abrupt and impossible to predict exactly. Switching state is much like switching to a different voice. Therefore, vibratory states are a source of confounding variation, for instance, when acquiring a voice range profile (VRP). In the quest for a state-aware, non-invasive VRP, a semi-automatic method based on the short-term spectrum of the electroglottographic signal (EGG) was developed. The method identifies rapid vibratory state transitions, such as the modal-falsetto switch, and clusters the EGG data based on their similarities in the relative levels and phases of the lower frequency components. Productions of known modal and falsetto voice were accurately clustered by a Gaussian mixture model. When mapped into the VRP, this EGG-based clustering revealed connected regions of different vibratory sub-regimes in both modal and falsetto.