|
|
|
| Custom Products | |
|
For many applications, very high quality speech can be produced by concatenating prerecorded pieces of speech. In this approach, a speaker records the most frequently occurring words, combinations of words, and phrase, and we store the collection of recordings in a sound inventory. To generate speech, we select the most appropriate piece of a word, whole word, word combination, or phrase. Because the speech quality of this type of system is very high, approaching natural human speech, we differentiate it from conventional speech synthesis systems, and call it a speech generation system. E-Speech has developed a speech generation system engine based on a proprietary unit selection and concatenation algorithm. We can produce a speech generation system for many different applications, voices, and languages. We will work with you to select a voice, determine an appropriate set of recordings, record the inventory and integrate the inventory into our speech generation engine. Our software can be used with minimal changes for many languages. We have successfully applied it to English and Spanish, and we used a similar system for Mandarin Chinese . Spanish Name Synthesis. Our Latin-American Spanish synthesis is a complete speech generation system that we have developed using the subword concatenation algorithms. Here are examples of other types of speech generation systems that E-Speech can develop for you. Flight Information. Here are some sample messages for a flight information system. In this system, input is a flight timetable as shown below, and our speech generation system produces the speech file in a female voice.
This approach is ideal for applications like names, addresses, driving directions, stock quotes, automated attendant, banking transactions, catalog fulfillment, airline reservations, travel schedules, credit card transactions, or insurance enrollment. People's names. Here are some examples of word concatenation for names in American English. In the following names, some of the first and last names were prerecorded as whole words, but many were produced by concatenation of subword-sized units.
For more detailed, technical information, please view the slides from our 1999 talk at the joint meeting of the Acoustical Society of America and the European Acoustical Society, Berlin, Germany. E-Speech can design and develop such a custom system for your application, including selection of words, word combinations, and phrases, choice of prosodic environments, and we can record and prepare the inventory, and we can develop the Unix or Win95/98/NT software to produce speech for your application. Please contact us at info@espeech.com to discuss how we could build a speech generation system for your application.
|
| © 1998-2007 E-Speech Corporation, Princeton, NJ. All Rights Reserved. |
