Norma Antonanzas‐Barroso

Integrated software for analysis and synthesis of voice quality.

Voice quality is an important perceptual cue in many disciplines, but knowledge of its nature is limited by a poor understanding of the relevant psychoacoustics. This article (aimed at researchers studying voice, speech, and vocal behavior) describes the UCLA voice synthesizer, software for voice analysis and synthesis designed to test hypotheses about the relationship between acoustic parameters and voice quality perception. The synthesizer provides experimenters with a useful tool for creating and modeling voice signals. In particular, it offers an integrated approach to voice analysis and synthesis and allows easy, precise, spectral-domain manipulations of the harmonic voice source. The synthesizer operates in near real time, using a parsimonious set of acoustic parameters for the voice source and vocal tract that a user can modify to accurately copy the quality of most normal and pathological voices. The software, user’s manual, and audio files may be downloaded from http:// brm.psychonomic-journals.org/content/supplemental. Future updates may be downloaded from www.surgery .medsch.ucla.edu/glottalaffairs/.

Perceptually motivated modeling of noise in pathological voices

Research investigating the correlation of acoustic measures of noise and the perception of pathological voice quality has consistently demonstrated a moderate association. However, this correlational approach cannot address basic questions concerning their cause and effect relationship, such as how carefully noise must be modeled to preserve the perceived pathological vocal quality, and what particular aspects of the noise are necessary to preserve natural vocal quality. To address these questions, an analysis by synthesis approach was applied to a large sample of natural pathological vowels varying in severity of pathology. Several synthetic versions of each voice were created using an LF source model and a formant synthesizer. Amplitude and spectral characteristics of the noise were varied as necessary. Stimuli were compared perceptually to the original voice and to each other. A discussion will focus on the procedures necessary to adequately model noise in voices with severe vocal pathology, and on the...

LF source model adequacy for pathological voices

The adequacy with which source models describe pathological phonation has been little studied, and the perceptual importance of deviations from a perfect model fit remains unknown. To address these issues, pathological voices were selected randomly from a library of samples. Glottal source waveforms were estimated by inverse filtering airflow signals. Formant frequencies and bandwidths were estimated using combined LPC and spectrographic analyses. Source functions were synthesized by least‐squares fitting a simplified LF model [Y. Qi and N. Bi, J. Acoust. Soc. Am. 96, 1182–1185 (1994)] to the glottal flow derivative. Three versions of each voice were created. The first was constructed by extracting one cycle from the original voice and repeating it to form a 1‐s signal. The second was constructed by recombining the inverse filtered source and estimated vocal tract resonances. The third version comprised the LF‐modeled source combined with the estimated vocal tract resonances. Expert listeners compared the...

Recent improvements to the University of California, Los Angeles' voice synthesizer

A number of enhancements have recently been added to the UCLA voice synthesizer. Additions have been made to add functionality, and to address several theoretical issues that arose during development and application. New functions include the ability to manipulate the source spectrum directly, by changing the slope and/or amplitude of a user-defined group of harmonics, or by manipulating individual harmonics directly. The synthesizer also allows users to add zeros to the vocal tract transfer function, improving modeling of many pathological voices. A number of other enhancements improve ability to create and systematically vary stimuli for perceptual experiments. Theoretical development has focused on the importance of pitch-synchronous Fourier analysis in modeling the voice source, particularly with respect to measuring and manipulating the amplitudes of H1 and H2. Issues surrounding spectral effects of upsampling, downsampling, and pulse stretching, which are needed for precise manipulation of F0, will ...

Synthesizer software for modeling voice quality

This poster presents a formant synthesizer that is designed especially for detailed modeling of voice quality. Users may interactively manipulate source pulse shapes and/or spectral characteristics of the voice source, the noise spectrum and noise‐to‐signal ratio, jitter and shimmer, vocal tremor rate and extent, and the vocal tract transfer function (formants and bandwidths), all in near‐real time. Pitch contours can be modeled directly or imported. The synthesizer may be utilized in method‐of‐adjustment tasks, or sequences of stimuli may be created for use in other experimental paradigms. An interactive audio‐visual display makes this a useful tool for teaching voice acoustics as well. Both an executable version of the program and the underlying code will be available during the conference. Copies of software that supports the voice modeling effort, including an interactive inverse filter, will also be distributed. [Work supported by NIDCD grant DC01797.]

The UCLA voice synthesizer, version 2.0

Based on our psychoacoustic model of voice quality, the UCLA voice synthesizer allows users to copy synthesize nearly any steady-state voice sample or to create stimuli that systematically vary in specific acoustic dimensions. This new release contains a number of significant improvements from earlier versions. The vocal tract model now includes three spectral zeros with adjustable bandwidths. The precise spectral shape of the harmonic and inharmonic sources can be modified at will, either by adjusting the heights of the harmonics or the noise amplitude in a selected frequency range or by specifying the desired spectral or noise slope in a range. Any number of ranges of any size can be specified. Synthesizer variables can be individually saved and copied between cases. Additional changes increase control of analysis parameters and the ease with which series of stimuli can be created. The synthesizer is fully documented and is freely available for download from headandnecksurgery.ucla.edu/glottalaffairs. C...

The effect of the spectral shape changes on voice perception

Researchers have long known that the shape of the vocal source spectrum is an important determinant of vocal quality, but the details regarding the importance of individual spectral features remains unclear. Previous research indicates four spectral features, H1‐H2, the spectral slope above 4 kHz, the slope from 1.5–2 kHz, and the slope from 2–4 kHz, account for virtually all the variability in spectral shapes. The present study provides preliminary evidence about the perceptual importance of these four features. Four series of stimuli were synthesized for each spectral parameter, in which that parameter varied in small steps. Because the perceptual salience of source parameters depends on F0 and on the spectrum of the inharmonic part of the source, series differed in the sex of the speaker (male/female) and in the NSR (noise‐free/very noisy). Listeners heard all possible pairs of voices within each series and were asked to determine whether stimuli were the same or different. We hypothesize that listener...

The effect of acoustic context and scale resolution on judgments of voice quality

Anchored protocols (in which listeners match test stimuli to fixed examples of different scale levels) control response variability in voice‐quality judgment tasks, and may be clinically applicable because they are also easy to use. Successful application requires that two conditions hold. First, listeners must be able to isolate the target quality in the natural pathologic voice and match the level of that quality to levels present in a different acoustic context (the anchors). Second, the anchored scales resolution must be sufficient for listeners to make consistent judgments about stimuli that fall between anchors. To examine these issues, listeners judged noise levels for 40 test voices in four tasks. Two were anchored tasks (limited‐scale resolution), in which the synthetic anchor voices matched the test voice (matched acoustic environment) in one condition and differed from the test voice environment in a second condition (traditional approach). Tasks three and four used the method of adjustment (fi...

Source modeling of severely pathological voices

Previous studies using a simplified LF source model [Qi and Bi, J. Acoust. Soc. Am. 96, 1182–1185 (1994)] to describe pathological voice sources indicated that model fit to the return phase is particularly poor. To examine the perceptual importance of differences in how the return phase is modeled, various functions were fitted to the source waveforms that had been obtained by inverse filtering. Voices were then resynthesized using these different source functions, and the resulting signals were compared perceptually to the original natural voice and to versions synthesized with LF modeled sources. The perceptual importance of changes in closing phase modeling for these severely pathological voices will be discussed, as will the range of functions necessary to obtain perceptually satisfactory synthesis of severely deviant voice sources. [Research supported by NIH.]