Generating words and names using N-grams of phonemes

• US 20070083369A1
• Filed: 10/06/2005
• Published: 04/12/2007
• Est. Priority Date: 10/06/2005
• Status: Active Grant

• Current Assignee: Daybreak Game Company, LLC (Enad Global 7 AB)
• Original Assignee: Sony Online Entertainment LLC (Enad Global 7 AB)
• Inventors: McCuller, Patrick
• Granted Patent: US 7,912,716 B2
• Time in Patent Office: Days
• US Class Current: 704/254
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams

Abstract

Generating words and/or names, comprising: receiving at least one corpus based on a given language; generating a plurality of N-grams of phonemes and a plurality of frequencies of occurrence using the corpus, such that each frequency of occurrence corresponds to a respective pair of phonemes and indicates the frequency of the second phoneme in the pair following the first phoneme in the pair; generating a phoneme tree using the plurality of N-grams of phonemes and the plurality of frequencies of occurrence; performing a random walk on the phoneme tree using the frequencies of occurrence to generate a sequence of phonemes; and mapping the sequence of phonemes into a sequence of graphemes.

Claims

1. A method for generating words and/or names, comprising:

receiving at least one corpus based on a given language;

generating a plurality of N-grams of phonemes and a plurality of frequencies of occurrence using said at least one corpus, such that each frequency of occurrence corresponds to a respective pair of phonemes and indicates the frequency of the second phoneme in the pair following the first phoneme in the pair;

generating a phoneme tree using said plurality of N-grams of phonemes and said plurality of frequencies of occurrence;

performing a random walk on said phoneme tree using said frequencies of occurrence to generate a sequence of phonemes; and

mapping said sequence of phonemes into a sequence of graphemes.

2. The method of claim 1, wherein said phoneme tree is configured with a root at the top and a plurality of nodes emanating from said root, said plurality of nodes connected in such a way that a pair of connected nodes are connected by a path from the first node in the pair to the second node in the pair,

wherein a pair of connected nodes corresponds to a pair of phonemes having a corresponding frequency of occurrence, so that the first node in a pair of nodes represents the first phoneme in the corresponding pair of phonemes, the second node in that pair of nodes represents the second phoneme in that pair of phonemes, and the path connecting the nodes of that pair of nodes represents the frequency of occurrence.

3. The method of claim 2, wherein said performing a random walk on said phoneme tree includes

selecting a set of connected paths through said phoneme tree by randomly selecting at a node a path from that node based on the frequency of occurrence for that path.

4. The method of claim 1, wherein said generating a plurality of N-grams of phonemes includes generating a plurality of N-gram frequency tables of phonemes.

5. The method of claim 1, wherein said generating a plurality of N-grams of phonemes includes generating a plurality of N-gram frequency trees of phonemes.

6. The method of claim 5, wherein generating a phoneme tree includes

combining said plurality of N-gram frequency trees of phonemes in a phoneme-to-phoneme connection.

7. The method of claim 1, further comprising

decomposing said at least one corpus into a sequence of words.

8. The method of claim 7, wherein said generating a plurality of N-grams of phonemes and a corresponding plurality of frequency of occurrences using said at least one corpus includes

decomposing said sequence of words into said sequence of phonemes.

9. The method of claim 8, wherein said decomposing said sequence of words includes

decomposing said sequence of words into said sequence of phonemes using a dictionary.

10. The method of claim 8, wherein said decomposing said sequence of words includes

decomposing said sequence of words into said sequence of phonemes using a grapheme-to-phoneme decomposition technique.

11. The method of claim 1, further comprising

determining whether said generated sequence of phonemes is pronounceable and reasonable to the speaker of said given language.

12. The method of claim 1, wherein said mapping said sequence of phonemes into a sequence of graphemes includes

transcribing said sequence of phonemes into said sequence of graphemes using a lookup table.

13. The method of claim 1, further comprising

forming words and/or names from said sequence of graphemes.

14. The method of claim 1, wherein said generating a plurality of N-grams of phonemes and a plurality of frequencies of occurrence using said at least one corpus and said generating a phoneme tree are configured to be substantially automatic without requiring user editing, modification, or other manual aid.

15. The method of claim 1, wherein said generating a plurality of N-grams of phonemes and a plurality of frequencies of occurrence using said at least one corpus and said generating a phoneme tree are configured to generate a unique phoneme tree, such that using a different corpus produces a different phoneme tree.

16. An apparatus for generating words and/or names, comprising:

means for receiving at least one corpus based on a given language;

first means for generating a plurality of N-grams of phonemes and a plurality of frequencies of occurrence using said at least one corpus, such that each frequency of occurrence corresponds to a respective pair of phonemes and indicates the frequency of the second phoneme in the pair following the first phoneme in the pair;

second means for generating a phoneme tree using said plurality of N-grams of phonemes and said plurality of frequencies of occurrence;

means for performing a random walk on said phoneme tree using said frequencies of occurrence to generate a sequence of phonemes; and

means for mapping said sequence of phonemes into a sequence of graphemes.

17. The apparatus of claim 16, wherein said phoneme tree is configured with a root at the top and a plurality of nodes emanating from said root, said plurality of nodes connected in such a way that a pair of connected nodes are connected by a path from the first node in the pair to the second node in the pair,

wherein a pair of connected nodes corresponds to a pair of phonemes having a corresponding frequency of occurrence, so that the first node in a pair of nodes represents the first phoneme in the corresponding pair of phonemes, the second node in that pair of nodes represents the second phoneme in that pair of phonemes, and the path connecting the nodes of that pair of nodes represents the frequency of occurrence.

18. The apparatus of claim 17, wherein said means for performing a random walk on said phoneme tree includes

means for selecting a set of connected paths through said phoneme tree by randomly selecting at a node a path from that node based on the frequency of occurrence for that path.

19. The apparatus of claim 16, wherein said first means for generating a plurality of N-grams of phonemes includes

third means for generating a plurality of N-gram frequency tables of phonemes.

20. The apparatus of claim 16, wherein said first means for generating a plurality of N-grams of phonemes includes

third means for generating a plurality of N-gram frequency trees of phonemes.

21. The apparatus of claim 20, wherein said second means for generating a phoneme tree includes

means for combining said plurality of N-gram frequency trees of phonemes in a phoneme-to-phoneme connection.

22. The apparatus of claim 16, further comprising

first means for decomposing said at least one corpus into a sequence of words.

23. The apparatus of claim 22, wherein said first means for generating a plurality of N-grams of phonemes and a corresponding plurality of frequency of occurrences using said at least one corpus includes

second means for decomposing said sequence of words into said sequence of phonemes.

24. The apparatus of claim 23, wherein said second means for decomposing said sequence of words includes

third means for decomposing said sequence of words into said sequence of phonemes using a dictionary.

25. The apparatus of claim 23, wherein said second means for decomposing said sequence of words includes

third means for decomposing said sequence of words into said sequence of phonemes using a grapheme-to-phoneme decomposition technique.

26. The apparatus of claim 16, wherein said means for performing a random walk on the phoneme tree includes

means for selecting a path to a following phoneme at a node of the phoneme tree from a plurality of paths based on said corresponding plurality of frequency of occurrences for said plurality of paths.

27. The apparatus of claim 16, further comprising

means for determining whether said generated sequence of phonemes is pronounceable and reasonable to the speaker of said given language.

28. The apparatus of claim 16, wherein said means for mapping said sequence of phonemes into a sequence of graphemes includes

means for transcribing said sequence of phonemes into said sequence of graphemes using a lookup table.

29. The apparatus of claim 16, further comprising

means for forming words and/or names from said sequence of graphemes.

30. The apparatus of claim 16, wherein said means for generating a plurality of N-grams of phonemes and a plurality of frequencies of occurrence using said at least one corpus and said means for generating a phoneme tree are configured to be substantially automatic without requiring user editing, modification, or other manual aid.

31. The apparatus of claim 16, wherein said means for generating a plurality of N-grams of phonemes and a plurality of frequencies of occurrence using said at least one corpus and said means for generating a phoneme tree are configured to generate a unique phoneme tree, such that using different corpus produces different phoneme tree.

32. An apparatus for generating words and/or names, comprising:

an input/output interface to receive at least one corpus based on a given language;

a processor configured to decompose said at least one corpus into a sequence of words, to generate a plurality of N-grams of phonemes and a plurality of frequencies of occurrence using said sequence of words and a dictionary, such that each frequency of occurrence corresponds to a respective pair of phonemes and indicates the frequency of the second phoneme in the pair following the first phoneme in the pair, and to generate a phoneme tree using said plurality of N-grams of phonemes and said plurality of frequencies of occurrence;

a storage for storing said dictionary, said phoneme tree, and a phoneme-to-grapheme lookup table,

wherein said phoneme tree is configured with a root at the top and a plurality of nodes emanating from said root, said plurality of nodes connected in such a way that a pair of connected nodes are connected by a path from the first node in the pair to the second node in the pair,

a pair of connected nodes in said phoneme tree corresponds to a pair of phonemes having a corresponding frequency of occurrence, so that the first node in a pair of nodes represents the first phoneme in the corresponding pair of phonemes, the second node in that pair of nodes represents the second phoneme in that pair of phonemes, and the path connecting the nodes of that pair of nodes represents the frequency of occurrence, and

said processor retrieving said phoneme tree to perform a random walk on said phoneme tree using said frequencies of occurrence to generate a sequence of phonemes, and maps said sequence of phonemes into a sequence of graphemes using said phoneme-to-grapheme lookup table.

33. A computer program, stored in a tangible storage medium, for generating words and/or names, the program comprising executable instructions that cause a computer to:

receive at least one corpus based on a given language;

generate a plurality of N-grams of phonemes and a plurality of frequencies of occurrence using said at least one corpus, such that each frequency of occurrence corresponds to a respective pair of phonemes and indicates the frequency of the second phoneme in the pair following the first phoneme in the pair;

generate a phoneme tree using said plurality of N-grams of phonemes and said plurality of frequencies of occurrence;

perform a random walk on said phoneme tree using said frequencies of occurrence to generate a sequence of phonemes; and

map said sequence of phonemes into a sequence of graphemes.

34. The computer program of claim 33, wherein said phoneme tree is configured with a root at the top and a plurality of nodes emanating from said root, said plurality of nodes connected in such a way that a pair of connected nodes are connected by a path from the first node in the pair to the second node in the pair,

wherein a pair of connected nodes corresponds to a pair of phonemes having a corresponding frequency of occurrence, so that the first node in a pair of nodes represents the first phoneme in the corresponding pair of phonemes, the second node in that pair of nodes represents the second phoneme in that pair of phonemes, and the path connecting the nodes of that pair of nodes represents the frequency of occurrence.

35. The computer program of claim 34, wherein executable instructions that cause a computer to perform a random walk on said phoneme tree includes executable instructions that cause a computer to

select a set of connected paths through said phoneme tree by randomly selecting at a node a path from that node based on the frequency of occurrence for that path.

36. The computer program of claim 33, wherein executable instructions that cause a computer to generate a plurality of N-grams of phonemes further includes executable instructions that cause a computer to

generate a plurality of N-gram frequency tables of phonemes.

37. The computer program of claim 33, wherein executable instructions that cause a computer to generate a plurality of N-grams of phonemes further includes executable instructions that cause a computer to

generate a plurality of N-gram frequency trees of phonemes.

38. The computer program of claim 37, wherein executable instructions that cause a computer to generate a phoneme tree further includes executable instructions that cause a computer to

combine said plurality of N-gram frequency trees of phonemes in a phoneme-to-phoneme connection.

39. The computer program of claim 33, further comprising executable instructions that cause a computer to

decompose said at least one corpus into a sequence of words.

40. The computer program of claim 39, wherein executable instructions that cause a computer to generate a plurality of N-grams of phonemes and a corresponding plurality of frequency of occurrences using said at least one corpus further includes executable instructions that cause a computer to

decompose said sequence of words into said sequence of phonemes.

41. The computer program of claim 40, wherein executable instructions that cause a computer to decompose said sequence of words further includes executable instructions that cause a computer to

decompose said sequence of words into said sequence of phonemes using a dictionary.

42. The computer program of claim 40, wherein executable instructions that cause a computer to decompose said sequence of words further includes executable instructions that cause a computer to

decompose said sequence of words into said sequence of phonemes using a grapheme-to-phoneme decomposition technique.

43. The computer program of claim 33, further comprising executable instructions that cause a computer to

determine whether said generated sequence of phonemes is pronounceable and reasonable to the speaker of said given language.

44. The computer program of claim 33, wherein executable instructions that cause a computer to map said sequence of phonemes into a sequence of graphemes further includes executable instructions that cause a computer to

transcribe said sequence of phonemes into said sequence of graphemes using a lookup table.

45. The computer program of claim 33, further comprising executable instructions that cause a computer to

form words and/or names from said sequence of graphemes.