Digital signal processing can create arbitrary signals, most of which have no relationship to those human beings
are capable of perceiving. With elaborate models of the human vocal tract's geometry, all sounds that humans can
generate can be created artificially.
Despite the universality of the technical infrastructure, Text to Speech (TTS) systems are usually designed
along national fault lines with localized voice fonts and linguistically identifiable entities; Sarahs for
US English, Heathers for UK English and Günthers for German.
It comes as no surprise that commercial TTS systems do not offer speech products with 'undesirable' features
such as slurred speech or, say, a strong German accent.
In order to query these cultural biases, I have crafted a German accented US English and a Mexican Spanish
accented US English system with limited vocabulary based on the SVOX speech engine.
>> Maria and Klara, synthetic information management agents, immigrate to the US and are hired as flight info providers at a busy airport.
Listen to Maria at work.
>>Klara's friend is taking an online astronomy course. But she is
confused about the planets of our solar system.
|
|
CURRENT PROJECTS : Make Language - Synthetic Accents