Search:
Menu VisualVoice / Step4

Visual Voice Home


About Artistic Vision People Teams Contact us


Activities



Publications Media Images/Movies


Opportunities


Related Links


Local Only


Website problems?

edit SideBar

Testing Phoneme Transitions

Before fully implementing the system, it is useful to test its intelligibility. An interface is created which allows the user to construct, store, and play back sequences of phoneme data. This is then plugged into sendToFace, the object from step 3. A visual inspection of the KuraFace suffices to determine the smoothness of transitions.

The interface, named morphTest, is shown below. It outputs sequences of phoneme data either by playing back a recorded file, or outputting values as they are created, and also allows the user to store values as they are created. The interface contains a coll( a storage object in MAX ) which can be filled with sequence values and cleared out as desired, and also supports file operations either to write the stored contents to a file, or to load files to the coll.

Creating Phoneme Streams

The options in the upper portion of the interface are primarily for constructing data sequences. They have the following behaviour:

i) initialize/reset values - resets all values, readies the system for a new sequence
ii) store values - a toggle switch, which if enabled will clear the coll, and place sequence steps there as they are created
iii) run - according to the "run/loop period" in milliseconds, will output a new sequence step once every period
iv) step - outputs sequence steps one at a time
v) print info - prints information on the current sequence step without outputting it or storing it to the coll
vi) hit with plosive - outputs a sequence step containing the plosive specified in the "plosive" text box.

Sequence Contents

Each "sequence step" is a list of values containing the following information:

i) phoneme probabilities - float between 0 and 1 for each vowel and consonant
ii) volume - float between 0 and 1 which is multiplied to each phoneme's probability
iii) vowel/consonant decision variable - float between 0 and 1 which determines relative weights of vowels & consonants
iv) plosive - character value for stopped consonants, "p k t ch b g d j" (usually set to "none" unless the plosive button is pressed).

Adjusting Sequence Values

Upon initialization, all values are reset to zero (or none for plosive). As the sequence is stepped or run, values can be modified in the following ways:

i) volume, vowel probability - adjusted by setting the values of their corresponding number boxes, in the center right of the interface. "increment v. prob" causes the vowel probability to change by that value at each step, up to the limits of 0 and 1.
ii) plosive - set using the "hit with plosive" button.
iii) phoneme probabilities - by typing the desired phoneme into the "first phoneme, second phoneme..." box and adjusting the corresponding multiplier, the phoneme's value changes by 1/100th the multiplier at each step. So, if upon initialization "F" is given a multiplier of 5, then at subsequent steps "F"'s probability will be 0.05, 0.1, 0.15... The phoneme's value can also be set, by typing it into "fourth phoneme" and choosing a value. Unless the phoneme is removed from this box, its value will be held at the value displayed.
iv) vowel probabilities - an x-y space is implemented which simulates the vowel space in the DIVA project. Users can load a profile, and accent, and if the interface is placed in "run" mode, then as the user grabs and moves the circle around the space, the corresponding vowel probabilities will be set, sequence steps with these values will be periodically sent out, effectively simulating vowels in performance mode.