Apparatuses and methods for training and operating speech recognition systems

• US 6,101,468 A
• Filed: 06/26/1997
• Issued: 08/08/2000
• Est. Priority Date: 11/13/1992
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Dragon Systems, Inc. (Microsoft Corporation)
• Inventors: Baker, James K., Heitman, Peter S., McGrath, Frank J., Sturtevant, Dean G., Steele, Elizabeth E., Parke, Joel W., Roberts, Jed M., Squires, Steven D., Gould, Joel M.
• Primary Examiner(s): Dorvil, Richemond
• Application Number:
  US08/882,914
• Time in Patent Office: 1,139 Days
• Field of Search: 704/251, 704/252, 704/253, 704/231, 704/235, 704/241, 704/244, 704/270, 704/275, 704/255, 704/243
• US Class Current: 704/251
• CPC Class Codes: G09B 19/04 Speaking with audible presentation of the material to be studied G09B5/04
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/285 Memory allocation or algorithm optimisation to reduce hardware requirements
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering
• CPC Class Codes: G10L 2015/0638 Interactive procedures
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A word recognition system can: respond to the input of a character string from a user by limiting the words it will recognize to words having a related, but not necessarily the same, string; score signals generated after a user has been prompted to generate a given word against words other than the prompted word to determine if the signal should be used to train the prompted word; vary the number of signals a user is prompted to generate to train a given word as a function of how well the training signals score against each other or prior models for the prompted word; create a new acoustic model of a phrase by concatenating prior acoustic models of the words in the phrase; obtain information from another program running on the same computer, such as its commands or the context of text being entered into it, and use that information to vary which words it can recognize; determine which program unit, such as an application program or dialog box, currently has input focus on its computer and create a vocabulary state associated with that program unit into which vocabulary words which will be made active when that program group has the focus can be put; detect the available computational resources and alter the instructions it executes in response; test if its ability to respond to voice input has been shut off without user confirmation, and, if so, turn that ability back on and prompt the user to confirm if that ability is to be turned off, store both a first and a second set of models for individual vocabulary words and enable a user to selectively cause the recognizer to disregard the second set of models for a selected word; and/or score a signal representing a given word against models for that word from different word model sets to select which model should be used for future recognition.

Claims

1. A computerized method of performing word recognition comprising the steps of:

receiving user generated word signals representing words to be recognized;

storing a plurality of vocabulary words, including storing, for each such word, pattern information identifying word signals which represent that word and a spelling, the letters of which can be either upper or lower case;

performing pattern matching between word signals and said pattern information to select which one or more of said vocabulary words most probably correspond to each such word signal; and

receiving user generated alphabetic signals in association with individual word signals; and

wherein said pattern matching includes the step of responding to the receipt of a given alphabetic signal in association with a given word signal by increasing the probability that the pattern matching will select as corresponding to that word signal a vocabulary word having an associated spelling which contains the alphabetic character represented by said given alphabetic signal, even though the case of that character in the vocabulary word'"'"'s associated spelling is not represented by the alphabetic signal.

2. A computerized method as in claim 1 wherein:

said receiving of user generated word signals includes receiving acoustic word signals representing the sounds of spoken words;

said pattern information stored for each vocabulary word includes information relating to the sounds of that word'"'"'s associated word signals; and

said pattern matching is speech recognition performed upon such acoustic word signals.

3. A computerized method as in claim 2 wherein:

said storing of vocabulary words includes storing character-representing vocabulary words each of which represents an alphabetic character; and

said receiving of user generated alphabetic signals includes;

receiving acoustic alphabetic signals, that is, acoustic signals representing the sounds of spoken utterances of said character-representing vocabulary words; and

causing said pattern matching to be performed upon the acoustic alphabetic signals to select which of said character-representing vocabulary words most probably corresponds to said acoustic alphabetic signal.

4. A computerized method as in claim 2 wherein said receiving of user generated alphabetic signals includes receiving such signal in the form of signals from a keyboard.

5. A computerized method as in claim 1 wherein said receiving of user generated alphabetic signals includes receiving alphabetic signals representing alphabetic characters of either upper or lower case.

6. A computerized method as in claim 5:

wherein said receiving of user generated alphabetic signals includes receiving in association with a given individual word signal a succession of one or more such signals each representing a characters which can be of either upper or lower case; and

said method further including;

storing a representation of the sequence of one or more characters represented by said succession of alphabetic signals, including the respective case of each such characters;

responding to certain conditions by producing as an output for the given word signal the spelling associated with one of said vocabulary word selected as most probably corresponding to a word signal; and

responding to certain other conditions by producing as an output for the given word signal said stored sequence of characters, including their respective case.

7. A computerized method as in claim 5 wherein said responding to the receipt of a given alphabetic signal by increasing the probability of selecting a vocabulary word having an associated spelling containing the alphabetic character represented by said given alphabetic signal includes:

selecting said vocabulary word from a class of words having said alphabetic character in either upper or lower case when said given alphabetic signal represents a lower case character; and

selecting said vocabulary word from a class of words having said alphabetic character only in the upper case when said given alphabetic signal represents an upper case character.

8. A computerized method as in claim 1 wherein:

said receiving of user generated alphabetic signals includes receiving a succession of one or more such signals, representing characters of either case, in association with a given individual word signal; and

said pattern matching includes;

performing said pattern matching to select an alternate choice group of one or more vocabulary words which most probably correspond to said given word signal; and

substantially limiting the words selected for the alternate choice group for said given word signal to words whose associated spellings start with the sequence of characters represented by said succession of alphabetic signals while not substantially limiting said selected words to words starting with the same sequence of cases as that of said sequence of characters.

9. A computerized method as in claim 8 further including:

displaying said sequence of characters represented by the alphabetic signals, if any, received so far for the given word signal;

displaying the words in said alternate choice group; and

allowing a user to select one of the words in said alternate choice group.

10. A computerized system for performing word recognition comprising:

elements receiving user generated word signals representing words to be recognized;

elements storing a plurality of vocabulary words, which store, for each such word, pattern information identifying word signals which represent that word and a spelling, the letters of which can be either upper or lower case;

elements performing pattern matching between word signals and said pattern information to select which one or more of said vocabulary words most probably correspond to each such word signal;

elements receiving a user generated string of one or more textual characters in association with an individual word signal; and

filtering elements responding to the receipt of said user generated string by causing said pattern matching means to select said vocabulary word from a class of such words having an associated spelling each of which contains a string of textual characters related to those of said user generated string, wherein said filtering elements includes logic requiring that each such related string be similar to the user generated string in certain ways, but need not be similar to it in other ways.

11. A computerized system as in claim 10 wherein said logic require that the characters in each such related string corresponding to certain characters, if any, in said user generated string match the letters represented by said certain characters, but need not match the case of those characters.

12. A computerized system as in claim 11 wherein said logic requires that character sequence of each such related string match the letters, but not the case, of lower case characters in the user generated string, and match both the letters and case of upper case characters in the user generated string.

13. A computerized system as in claim 10 wherein said logic requires said related strings to correspond to the start of the spelling of their respective associated vocabulary word'"'"'s pronunciation, and does not require said related strings to correspond to the start of said respective vocabulary word'"'"'s spelling when said spelling includes characters before the start of the spelling of said pronunciation.