[Alicia Wassink's homepage | UW Sociolinguistics Lab | VOIS3D.htm]
VOIS3D (Vowel Overlap Indication Software - 3D)
Contents:
1. What is VOIS3D?
2. Running the Standalone Application
3. Project Summary
4. References and citation information
version information: (.exe) PC only [MAC users can run VOIS3D in a virtual Windows operating system environment, such as VMWare or Parallels]
1. What is VOIS3D?
VOIS3D is a software tool for acoustic phonetic, sociolinguistic, and experimental phonological research. VOIS3D integrates previous and new methods for vowel formant frequency normalization and duration normalization, with a new method for graphical vowel representation that allows numerical. This is possible because VOIS may take as its input either previously normalized vowel data, or raw data which may then be normalized as part of the graphing process.
Big idea: Normalized scatter for two vowel distributions is modeled as two best-fit ellipses oriented at angles with respect to F1, F2 axes. The output of the metric is an overlap fraction. The overlap fraction represents the area of the region of overlap (the region shared by both best-fit ellipses).
Procedure:
(1) Each observed vowel is plotted in a coordinate system where (x,y) = (normF1, normF2)
(2) Center each vowel class' datapoints around its own origin. (0,0) is the center of each system's vowel space. Determine the geometric formula for each ellipse and define a principal axis for each.
(3) Rotate each ellipse along its principal axis; determine the range of coordinates it occupies within its own 2D space.
(4) Using area information, determine extent of overlap between areas of Vowel 1 & Vowel 2. The output of this procedure is referred to as the "overlap fraction", a real-number value between 0-100%. (Wassink 1999,2006)
Visualization functions in VOIS3D:
(a) Scatterplot view (user's raw data).
(b) Two-dimensional overlap for raw or parametric data (parametric data shown).
(c) Ellipse view (2-dimensional view of parametric data). Raw scatter may be plotted with parametric ellipses (2-D) and ellipsoids (3-D) (shown in program screenshot, below).
(d) Three-dimensional overlap view for raw or parametric data (parametric data shown). Rotation of the cube is possible with the full version of VOIS3D running under MATLAB.
2. Running the Standalone Application
Note: For installation of VOIS3D standalone software. Matlab software not required for operation.
Files you need to install VOIS3D (PC):
1. mglinstaller.exe (creates the C:\MATLAB6p5p1\bin\win32 directory)
2. VOIS3D.zip
3. download the manual: (pdf)
Where to install them:
1. Unzip the VOIS3D.zip file. This should create VOIS3D.exe and bin folders
2. Copy all files with the .dll extension (*.dll) and drop the copies into the "matlab/bin/win32" folder (make sure they keep their original names)
3. Also copy the VOIS3D.exe file and drop the copy into the matlab/bin/win32 folder
4. Copy any files from the bin folder that was unzipped with the .exe file into the matlab/bin/win32 folder. It is ESSENTIAL for the VOIS3D.exe file to be in the matlab/bin/win32 folder for the standalone application to work.
5. Because you are running from copies of the original files, the original files will be available to you in case you accidentally delete a file, or need to redo this process.
Launching VOIS3D:
Double-click on the VOIS3D.exe program file. A dos screen may appear first, but don't worry. The VOIS3D window should soon appear in the foreground.
Running VOIS3D:
For help beginning your VOIS3D session, consult the file "VOIS3D Manual".
3. project summary
Project Title: Vowel Overlap Indication Software (VOIS3D) Project,
Funding: University of Washington, RRF Grant 65-2597, Proposal #2952
Budget Period: 2/1/03 to 1/31/04
Alicia Beckford Wassink, Department of Linguistics
submitted: 6 May, 2004
Scope of Work
This project involved the development of computer software for use by researchers in several subfields of linguistics, including sociolinguistics, phonology, and phonetics. The project had three phases: first, modification of the preexisting computer program (Wassink, 1999) that provided an analytic geometric solution for assessment of spectral overlap (Spectral Overlap Assessment Metric) for two spectral dimensions (F1, F2). The initial modification involved provision of a graphical user interface and extension of the program's functionality to assess overlap in either two or three dimensions; second, construction of three sets of appropriate linguistic data for the purpose of testing and demonstrating application of the software to topics of theoretical interest. Finally, the proposed research was to culminate in the submission of a scholarly article for publication (Wassink, 2006). The specific outcomes of this RRF-funded project are the extension of the user interface and development of a set of test data. These are described below.
Extension of the User Interface
One major undertaking of the project was to find a suitable programming environment to use in extending the existing program (written in MATLAB 4.2c.1). A number of programming environments were considered and tested, including .NET, Flash, Visual Studio C++, C-sharp, and Open GL. In the end, it was possible to extend the user interface and accomplish necessary programming work using MATLAB 6.5.1 software. This was quite advantageous because it did not require rewriting the core code for the overlap algorithm. MATLAB 6.5.1 offers advanced graphical tools such as user input dialogue boxes, buttons, object handles, etc. that greatly improve the accessibility and usability of the program (i.e., the user need not type in line-level commands). It is cross-platform compilable (for use on Macs and PCs). In addition, this newest version of MATLAB enables the target data to be read into the program from several types of files (tab-delimited text, comma-delimited text, Microsoft Excel databases), or directly input to the program. This further simplifies the user's tasks. He or she may select among external data files using a menubar, rather than having to type the name of the source files into the VOIS3D code itself, as was required in the previous version of the program. The figure below shows the new interface and a sample output window.
An unexpected bit of functionality incorporated at no additional cost (but which had not been planned in the project proposal) is that the new VOIS3Dprogram is able to normalize spectral and durational data using several different algorithms already used in the discipline (e.g., Lobanov, 1971; Nearey, 1977; Shirai, 2004). Previously, the software required that data be normalized before use of the program. However, it became evident that end-users might utilize VOIS3Dto compare overlap among different types of acoustic data, e.g. within- or between-speaker comparisons, or within-speaker comparisons of time series or other paired data. A means was devised whereby the program automatically checks the composition of the user's database and finds possible grouping variables along which the user may partition their data. The VOIS3D manual informs the user of the normalization procedures available in VOIS3Dand data types most appropriate for each, and the user chooses a method for normalization from a pulldown menu. All eligible methods may then be executed, and their results subjected to side-by-side comparison in a results field on the main screen. The user can also specify when the data to be checked for overlap were already normalized before being imported into VOIS3D.
Impact of the award
This RRF award enabled the development of a computer program that both assesses spectral and temporal overlap, and provides automatic normalization. The novel mathematical algorithm that this program uses is now available to linguists in several fields. The normalization functionality of the program significantly broadens the potential usefulness of the VOIS3Dsoftware. Even if users do not use the overlap calculation functions, these capabilities make VOIS3D a valuable tool to phoneticians and sociolinguists, who often resort to entering formulae into Microsoft Excel in order to normalize data.
Acknowledgments: Many thanks to Jeremy Waltmunson, Setsuko Shirai and Dirk Beckford Wassink.
**to cite the VOIS3D program, please cite the Wassink (2006) reference, below**.
Lobanov, B. M. (1971). Classification of Russian Vowels Spoken by Different Speakers, J. Acoust. Soc. Am. 49(3), pp. 606-608.
Nearey, T. M. (1977). Applications of generalized linear modeling to vowel data, Proceedings ICLSP 92, pp. 583-586.
Shirai, S. (2004). Lexical effects in Japanese vowel reduction, Unpublished PhD. thesis, University of Washington.
Wassink, A. B. (2006) A geometric representation of spectral and temporal vowel features: Quantification of vowel overlap in three varieties, Journal of the Acoustical Society of America, 119(4), pp 2334-2350.
Wassink, A. B. (1999) A sociophonetic analysis of Jamaican vowels. Unpublished PhD dissertation, University of Michigan, Ann Arbor.
© 1999-2009 Alicia Beckford Wassink. Copyright managed by University of Washington Office of Intellectual Property and Technology Transfer.