Voice-optimized database system and method of using same

• US 6,377,927 B1
• Filed: 10/07/1998
• Issued: 04/23/2002
• Est. Priority Date: 10/07/1998
• Status: Expired due to Term

• Current Assignee: LogicTree, LLC
• Original Assignee: Fred F. Korangy, Masoud Loghmani
• Inventors: Loghmani, Masoud, Korangy, Fred F.
• Primary Examiner(s): Korzuch, William
• Assistant Examiner(s): MCFADDEN, SUSAN IRIS
• Application Number:
  US09/167,486
• Time in Patent Office: 1,294 Days
• Field of Search: 704/275, 704/270, 704/256, 704/231, 704/246, 704/257
• US Class Current: 704/275
• CPC Class Codes: G06F 16/243 Natural language query formulation
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

A voice-optimized database system and method of audio vector valuation (AVV) provides means to search a database in response to a spoken query. Audio vectors (AVs) are assigned to phonemes in the names or phrases constituting searchable items in a voice-optimized database and in spoken queries. Multiple AVs can be stored which correspond to different pronunciations of the same searchable items to allow for less precision on the part telephone callers when stating their queries. A distance calculation is performed between the audio vectors of database items and spoken queries to produce search results. Existing databases can be enhanced with AVV. Several alternate samples of a spoken query are generated by analyzing the phonemic string of the spoken query to create similar, alternate phonemic strings. The phonemic string of the spoken query and the alternate phonemic strings are converted to text and used to search the database.

Claims

1. A voice-optinized database system for allowing users to access an electronic shopping cart to purchase items offered through an online shop using spoken queries comprising:

a memory device for storing records for said items; and

a processing device connected to said memory device and operable to retrieve selected ones of said records from said database in response to said spoken queries, said memory device being configured to store for each of said records at least one phoneme vector for parsing of said memory device by said processing device in response to said spoken queries;

wherein each of said records has at least one searchable field comprising item data that is characterized by phonemes, said phonemes being assigned respective values, said phonemes having similar pronunciation being assigned similar values, each said phoneme vector comprising said values corresponding to said phonemes in said at least one searchable field.

2. A database system as claimed in claim 1, wherein each said phoneme vector comprises vector components, one of said vector components indicating the number of phonemes associated with said item data in the corresponding said at least one searchable field, and other said vector components comprising respective said values of said phonemes in the corresponding said item data.

3. A method of retrieving items from a database in response to a spoken query comprising the steps of:

assigning a database item phoneme vector to each of a plurality of items in said database, each of said items comprising at least one phoneme, each said phoneme having a unique value assigned thereto and stored in a table, said database item phoneme vector comprising a vector component indicating the number of said phoneme of which the corresponding said item is composed, and phoneme vector components which each comprise said unique value corresponding to a respective said phoneme in one of said items;

determining a spoken phrase phoneme vector for said spoken query, said spoken query being analyzed to determine each said phoneme thereof, said spoken phrase phoneme vector of said spoken query comprising said phoneme vector components each having said unique value for a respective said phoneme in said spoken query and a value indicating the number of said phoneme in said spoken query; and

comparing at least a portion of said spoken phrase phoneme vector with at least a portion of said database item phoneme vector corresponding to each of said items.

6. A method as claimed in claim 3, wherein a plurality of said phoneme having similar pronunciation are assigned unique values which are numerically close in value and a plurality of said phoneme having distinct pronunciation are assigned unique values having increasingly different numerical values depending on the degree of distinctness in pronunciation.

7. A method as claimed in claim 6, wherein said unique value is a vector in a matrix having a number of axes corresponding to a number of levels of contrast desired between said phoneme having similar pronunciation.

8. A method as claimed in claim 3, wherein said database comprises context data to relate groups of said item s to different subjects, said comparing step comprising the step of:

determining to which of said subjects said spoken query relates; and

comparing at least a portion of said spoken phrase phoneme vector with at least a portion of said database item phoneme vector corresponding to each of said items relating to said subject.

9. A method as claimed in claim 3, where in said database comprises records having data fields and context data to relate at least one of said records and said data fields to different subjects, said comparing step comprising the step of:

determining to which of said subjects said spoken query relates; and

comparing at least a portion of said spoken phrase phoneme vector with at least a portion of said database item phoneme vector corresponding to each of at least one of said records and said data fields relating to said subject.

10. A method as claimed in claim 3, further comprising the steps of:

selecting multiple pronunciations of each of said items;

assigning a corresponding said database item phoneme vector to each of said multiple pronunciations for storage in said database for the corresponding one of said items; and

comparing at least a portion of said spoken phrase phoneme vector with at least a portion of each said database item phoneme vector corresponding to each of said items and said corresponding multiple pronunciations of said items.

4. A voice-optimized database system for allowing users to access an electronic shopping cart to purchase items offered through an online shop using spoken queries comprising:

a memory device for storing records for said items; and

a processing device connected to said memory device and operable to retrieve selected ones of said records from said database in response to said spoken queries, said memory device being configured to store for each of said records at least one audio vector for parsing of said memory device by said processing device in response to said spoken queries;

wherein each of said records has at least one searchable field comprising item data that is characterized by phonemes, said phonemes being assigned respective values, said phonemes having similar pronunciation being assigned similar values, each said audio vector comprising said values corresponding to said phonemes in said at least one searchable field;

wherein said comparing step comprises the steps of

assigning query coordinates to respective said phoneme represented in at least a portion of said spoken phrase audio vector to define said portion of said spoken phrase audio vector in accordance with a matrix position in a matrix having as many axes as the number of said phoneme in said portion of said spoken phrase audio vector, each of said query coordinates having a value corresponding to said unique value of the related said phoneme;

assigning item coordinates to respective said phoneme in at least a portion of each said database item audio vector to define said portion of each said database item audio vector in accordance with a matrix position in said matrix, said portion of said database item audio vector as having the same number of said phoneme as said portion of said spoken phrase audio vector, each of said items coordinates having a value corresponding to said unique value of the related said phoneme; and

calculating distance between respective ones of said query coordinates and said item coordinates to determine which of said items are similar to said spoken query.

5. A method as claimed in claim 14, wherein said calculating step comprises the steps of:

removing the most distal ones of said item coordinates from said database item audio vector when said spoken phrase audio vector has fewer said phoneme vector components than said database item audio vector; and

removing the most distal ones of said query coordinates from said spoken phrase audio vector when said database item audio vector has fewer said phoneme vector components than said spoken phrase audio vector.

11. A method of providing an existing database with a spoken query response capability comprising the steps of:

accessing a library of phonemes;

translating a spoken query into a group of said phonemes corresponding thereto and selected from said library;

converting said group of phonemes into query text;

generating alternate test by performing at least one of a plurality of operations comprising processing said group of phonemes to produce an alternate group of phonemes by selectively using at least one of individual said phonemes, divided said phonemes and regrouped said phonemes and converting said alternate group of phonemes to text, and selecting other text that is related said query text, said other text characterized by at least one of similar pronunciation to any of said group of phonemes, comprising allomorphs similar to allomorphs in any of said group of phonemes and having similar meaning to said query text but different format; and

conducting searches of said database in response to said spoken query using said query text and said alternate text.

12. A method as claimed in claim 11, wherein said phonemes in said library are assigned respective values, said values of said phonemes having similar pronunciation being close numerically and said values of said phonemes having distinct pronunciation having greater numerical differences than said values of similar said phonemes, said selecting step comprising the step of selecting said other text having said phonemes with said values that are numerically close with respect to said values of any of said group of phonemes.

13. A method as claimed in claim 11, wherein said query text corresponds to a number and said selecting step comprises the step of selecting said other text corresponding to names of digits in said number, and names of decimal components of said number.