System, method, and technique for identifying a spoken utterance as a member of a list of known items allowing for variations in the form of the utterance

• US 7,725,309 B2
• Filed: 06/06/2006
• Issued: 05/25/2010
• Est. Priority Date: 06/06/2005
• Status: Active Grant

• Current Assignee: Apple Inc.
• Original Assignee: Novauris Technologies Ltd. (Apple Inc.)
• Inventors: Bedworth, Mark
• Primary Examiner(s): MCFADDEN, SUSAN IRIS
• Application Number:
  US11/422,406
• Publication Number:
  US 20060277030A1
• Time in Patent Office: 1,449 Days
• Field of Search: 704/243
• US Class Current: 704/10
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

An application-specific dictionary is provided that includes a plurality of entries, where each entry matches to one or more items that comprise one or more records of a record store. For at least some of the plurality of entries, each entry has multiple phonetic representations, including one or more phonetic representations that are at least one of (i) an abbreviation of that entry, or (ii) a null use of the entry when that entry is combined with another entry.

Claims

1. A method for performing speech recognition, the method being performed by one or more processors that perform steps comprising:

identifying a plurality of entries from a sequence of phonetic units that are recognized from a spoken input, wherein the plurality of entries individually match to one or more items that comprise one or more records of a database that comprises a plurality of records;

accessing a search node for the database, wherein the search node includes, for each record of at least some of the plurality of records, multiple phonetic representations, including at least one phonetic representation that contains at least one of (i) an abbreviation of an entry that is included in the record, or (ii) a null use of an entry that is included in the record when that entry is combined with one or more other entries; and

selecting, from the search node, a matching record of the database for the spoken input, wherein selecting the matching record includes comparing the plurality of entries identified from the spoken input to the phonetic representations of the individual records, including to the multiple phonetic representations of records in the at least some of the plurality of records.

2. The method of claim 1, identifying a plurality of entries from a sequence of phonetic units that are recognized from a spoken input includes identifying one or more names, and wherein accessing the search node includes accessing the search node for a directory of people and/or businesses.

3. The method of claim 2, wherein identifying a plurality of entries includes identifying a nickname or alias as one of the plurality of entries that are recognized from the spoken input, and wherein selecting a matching record further comprises selecting a given record in which at least one of multiple phonetic representations includes the entry for the nickname or alias, and wherein the nickname of alias is orthographically dissimilar to a corresponding name contained in the matching record.

4. The method of claim 1, wherein identifying a plurality of entries from a sequence of phonetic units includes ignoring an entry corresponding to a title of a person.

7. The method of claim 1, wherein accessing a search node includes accessing the search node for a media library of audio or visual works, and wherein the plurality of records that comprise the media library include one or more entries selected from a group consisting of (i) song title, (ii) artist name, (iii) genre of audio visual work, and (iv) file name.

5. The method of 1, wherein selecting the matching record includes selecting the matching record in which two or more of the entries that comprise the record have a canonical arrangement that is different than a sequence of spoken entries that are determined to correspond to the two or more entries.

6. The method of claim 5, wherein the sequence of entries that are determined to correspond to the two or more entries include an abbreviation of one of the two or more entries.

8. A computer-readable medium that stores instructions for performing speech recognition, the instructions including a set of instructions, that when executed by the one or more processors, cause the one or more processors to perform steps comprising:

identifying a plurality of entries from a sequence of phonetic units that are recognized from a spoken input, wherein the plurality of entries individually match to one or more items that comprise one or more records of a database that comprises a plurality of records;

accessing a search node for the database, wherein the search node includes, for each record of at least some of the plurality of records, multiple phonetic representations, including at least one phonetic representation that contains at least one of (i) an abbreviation of an entry that is included in the record, or (ii) a null use of an entry that is included in the record when that entry is combined with one or more other entries; and

selecting, from the search node, a matching record of the database for the spoken input, wherein selecting the matching record includes comparing the plurality of entries identified from the spoken input to the phonetic representations of the individual records, including to the multiple phonetic representations of records in the at least some of the plurality of records.