NonText Version

Speech Research Laboratory
AI. duPont Hospital for Children
and the
University of Delaware

SRL Main Page (text)

Info for Users (text)
Info for Clinicians (text)
Info for Researchers (text)

ModelTalker Speech Synthesizer

STAR: Speech Training, Assessment and Remediation (text)

Language and Speech

Newsletter (text)

Contact Us!


Info for Users

Index:

How a Synthetic Voice is Created

The first step in creating a synthetic voice is recording the inventory of words and phrases. A software package called InvTool will prompt the person being recorded with a specific word or phrase. Guided by the InvTool instructions, the person will say and store the prompted word or phrase. InvTool will give feedback about the quality of the recording and may make suggestions to the person about the volume, pitch, and goodness of pronunciation, and the person can choose to rerecord the recording or save it. Once the person is satisfied, she or he pushes a button to store the speech and InvTool prompts for the next word or phrase. If the person needs to take a break, InvTool will remember which word they were on and, when the person is ready, will resume where the person left off.

Currently, InvTool will prompt for approximately 1400 words and phrases. (However, our lab is still in the process of investigating ways of reducing the size of the inventory.)

Once the inventory has been recorded, InvTool will run the database of recorded speech through another software program called BCC. BCC converts the recorded speech into a form that the ModelTalker speech synthesizer can use. BCC may report errors, which may make it necessary for the person who recorded the inventory to rerecord certain words and phrases. Once BCC has finished processing the recorded inventory, the processed database is ready for use by the ModelTalker synthesizer.

The ModelTalker synthesizer can be thought of as consisting of two components: a user interface; and the speech synthesizer. The user interface is what a person sees on the computer screen. The person types in text or enters text into the synthesizer from a file, and the text goes through the user interface to the synthesizer, which creates synthetic speech from the database. If a person uses an AAC device and would prefer using that to using ModelTalker's user interface, ModelTalker's speech synthesizer can be put into most computer-based AAC devices. Thus, an AAC device user can continue to use the device he or she is familiar with, only with a personalized voice. (To further simplify the substitution process, we will be changing aspects of the ModelTalker synthesizer to make it fully compliant with emerging standards in speech synthesizer applications.)