1. Field of the Invention
The present invention relates to a method of training a computer system via human voice input from a human teacher, with the computer system including a speech recognition engine.
2. Background Art
A large concatenated voice system with a large vocabulary is capable of speaking a number of different words. For each word in the vocabulary of the large concatenated voice system, the system has been trained so that a particular word has a corresponding phonetic sequence. In large concatenated voice systems and other so-called artificial intelligence systems, manual data entry is usually used to train the systems. This is usually done by first training a data entry person the advanced skill sets required to program the phonetic knowledge into specific elements of the computer program for storage and future use. This type of training technique is tedious, prone to errors, and has a tendency to be academic in entry style rather than capturing a true example of how a word is pronounced or what a word, phrase, or sentence means or translates to.
Although the use of manual data entry to train large concatenated voice systems has been used in many applications that have been commercially successful, manual data entry training techniques have some shortcomings. As such, there is a need for a method of training a computer system that overcomes the shortcomings of the prior art.
It is, therefore, an object of the present invention to provide a method of training a computer system via human voice input from a human teacher.
In carrying out the above object, a method of training a computer system via human voice input from a human teacher is provided. The computer system has a text to speech engine and a speech recognition engine. The method comprises presenting a text spelling of an unknown word, and receiving a human voice pronunciation of the unknown word from the human teacher. The method further comprises determining a phonetic spelling of the unknown word with the speech recognition engine based on the human voice pronunciation of the unknown word. The text spelling is associated with the phonetic spelling to allow the text to speech engine to correctly pronounce the unknown word in the future, when presented with the text spelling of the unknown word.
It is appreciated that the phonetic spelling determined for the unknown word with the speech recognition engine may include a sequence of phonemes names and/or known words. In a preferred embodiment, after presenting the text spelling of the unknown word, the computer system, using speech output, requests to receive the human voice pronunciation of the unknown word. The request from the computer system takes a form of an ongoing dialog between the computer system and the human teacher. More preferably, the method further comprises establishing a plurality of request statements. Each request statement has an information content level. The information content levels range from a low information content level to a high information content level. The plurality of request statements are used by the computer system during the ongoing dialog. Most preferably, presenting, receiving, determining, and associating are repeated for a plurality of unknown words. The information content level for the request statements in the ongoing dialog progressively lessens as presenting, receiving, determining, and associating are repeated.
Further, in carrying out the present invention, a method of training a computer system via human voice input from a human teacher is provided. The computer system has a speech recognition engine. The method comprises receiving a human voice pronunciation of an unknown word from the human teacher. The method further comprises determining a phonetic spelling of the unknown word with the speech recognition engine based on the human voice pronunciation of the unknown word, and receiving a known word that is related in meaning to the unknown word. The known word is associated with the phonetic spelling of the unknown word to allow the speech recognition engine to correctly recognize the unknown word in the future as related in meaning to the known word.
Preferably, receiving the known word further comprises receiving a human voice pronunciation of the known word from the human teacher. Alternatively, receiving the known word further comprises receiving a text spelling of the known word.
Still further, in carrying out the present invention, a computer readable storage medium having instructions stored thereon that direct a computer to perform a method of training a computer system via human voice input from a human teacher is provided. The computer system has a text to speech engine and a speech recognition engine. The medium further comprises instructions for presenting a text spelling of an unknown word, and instructions for receiving a human voice pronunciation of the unknown word from the human teacher. The medium further comprises instructions for determining a phonetic spelling of the unknown word with the speech recognition engine based on the human voice pronunciation of the unknown word. And further, the medium further comprises instructions for associating the text spelling with the phonetic spelling. This association allows the text to speech engine to correctly pronounce the unknown word in the future when presented with the text spelling of the unknown word.
Even further, in carrying out the present invention, a computer readable storage medium having instructions stored thereon that direct a computer to perform a method of training a computer system via human voice input from a human teacher is provided. The computer system has a speech recognition engine. The medium further comprises instructions for receiving a human voice pronunciation of an unknown word from the human teacher, and instructions for determining a phonetic spelling of the unknown word with the speech recognition engine based on the human voice pronunciation of the unknown word. The medium further comprises instructions for receiving a known word that is related in meaning to the unknown word, and instructions for associating the known word with the phonetic spelling of the unknown word. The association allows the speech recognition engine to correctly recognize the unknown word in the future as related in meaning to the known word.
The advantages associated with embodiments of the present invention are numerous. In accordance with the present invention, a system and method to train computer systems via human voice input are provided. Automatic phonetic transcription may be used to enable human teaching of semi-intelligent computer systems correct pronunciation for speech output and word, phrase, and sentence meanings. Further, speech output from and human speech input to a computer may be used to ask human teachers questions and accept input from the human teacher to improve performance of the computer system.
The above object and other objects, features, and advantages of the present invention will be readily appreciated by one of ordinary skill in the art in the following detailed description of the preferred embodiment when taken in connection with the accompanying drawings.
With reference now to
In accordance with the present invention, computer system 10 is trained via human voice input from a human teacher. First, computer 12 is presented with a text spelling of an unknown word. The text spelling of the unknown word may be presented to computer 12 in a variety of ways. For example, computer 12 may manually receive the text spelling of the unknown word, or may, in any other way, come across the text spelling of the unknown word. Thereafter, a human voice pronunciation of the unknown word is received by system 10 at microphone 24 from a human teacher. Speech recognition engine 16 determines a phonetic spelling of the unknown word based on the human voice pronunciation of the unknown word. It is appreciated that the phonetic spelling may include a sequence of phonemes names and/or known words as determined by word recognizer 18 and/or database with phonetics 20. Further, in a preferred implementation, after the text spelling of the unknown word is presented, system 10, using speech output at speaker 22, requests to receive the human voice pronunciation of the known word.
In a preferred embodiment, the request by the computer system to receive the human voice pronunciation of the unknown word takes a form of an ongoing dialog between the computer system and the human teacher as illustrated by example in
That is, in accordance with the present invention, speech output from and speech input to a computer is used to ask human teachers questions and accept input from the human teacher to improve performance of the computer system. The improved performance can be: how the computer is performing an operation such as pronouncing a word or assembling a sentence or phrase, or how the computer is translating information. A natural dialog with the computer can be set so that realistic data can be captured. For example, if the word “bozotron” is being pronounced by the system, the computer can ask the teacher for advice on how to pronounce the word. The computer would have a list of ways to ask the questions with a variable for the questionable data. Further, the computer may develop its own questions.
As best shown in
It is appreciated that a plurality of statements are established for use by the computer during the dialog with the human teacher. In a preferred implementation, each statement or request statement (because the statements are used to ultimately request to receive the human voice pronunciation of the unknown word from the human teacher) has an information content level. The information content levels range from a low information content level to a high information content level. The plurality of request statements are used by the computer system during the ongoing dialog.
Preferably, during the ongoing dialog, the computer system progressively lessens the information content level for the request statements used in the ongoing dialog. For example, at block 32, the computer may explain that it has several words that it does not know how to pronounce. Thereafter, for the first unknown word, request statements having high information content levels are used until the text spelling of the unknown word is associated with a phonetic spelling. Thereafter, the computer system may repeat the same steps, this time for the second unknown word, but this time using request statements having a slightly lower information content level. And again, after the second unknown word text spelling has been associated with a phonetic spelling, the process may again be repeated for the third word. This time, for the third word, an even lower information content level may be used for the request statements. The use of progressively lower information content levels for the request statements provides a more natural conversation flow between the human teacher and the computer system. For example, by the time the computer is asking to receive the human voice pronunciation of a tenth word, it is no longer necessary for the computer to say “I have a new word that I do not know how to pronounce. Do you have time to listen to my question?” Instead, the computer may say “Want to hear the next one?” or “Got time for another?”
It is appreciated that embodiments of the present invention provide a method of training a computer system via human voice input from a human teacher. Automatic phonetic transcription is used to enable human teaching of semi-intelligent computer systems correct pronunciation for speech output and word, phrase, and sentence meanings. As shown in
Another embodiment of the present invention is illustrated in
It is appreciated that in the method illustrated in
It is appreciated that in accordance with the present invention, methods may be implemented via a computer readable storage medium having instructions stored thereon that direct a computer to perform a method of the present invention. That is, the methods as described in
In addition, the method illustrated in
In addition, it is appreciated that all optional features and preferred features described herein for methods of the present invention may also be implemented as instructions on a computer readable storage medium.
While embodiments of the invention have been illustrated and described, it is not intended that these embodiments illustrate and describe all possible forms of the invention. Rather, the words used in the specification are words of description rather than limitation, and it is understood that various changes may be made without departing from the spirit and scope of the invention.
Number | Name | Date | Kind |
---|---|---|---|
5682539 | Conrad et al. | Oct 1997 | A |
5724481 | Garberg et al. | Mar 1998 | A |
5852801 | Hon et al. | Dec 1998 | A |
6041300 | Ittycheriah et al. | Mar 2000 | A |
6078885 | Beutnagel | Jun 2000 | A |
6092044 | Baker et al. | Jul 2000 | A |
6125341 | Raud et al. | Sep 2000 | A |
6144938 | Surace et al. | Nov 2000 | A |
6233553 | Contolini et al. | May 2001 | B1 |
6321196 | Franceschi | Nov 2001 | B1 |
6411932 | Molnar et al. | Jun 2002 | B1 |
6598018 | Junqua | Jul 2003 | B1 |
6598020 | Kleindienst et al. | Jul 2003 | B1 |
6629071 | Mann | Sep 2003 | B1 |
6694296 | Alleva et al. | Feb 2004 | B1 |
6721706 | Strubbe et al. | Apr 2004 | B1 |
6823313 | Yuchimiuk et al. | Nov 2004 | B1 |
20020055844 | L'Esperance et al. | May 2002 | A1 |
20030182111 | Handal et al. | Sep 2003 | A1 |
Number | Date | Country | |
---|---|---|---|
20030130847 A1 | Jul 2003 | US |