Apparatus and methods for pronunciation lexicon compression

• US 20060004564A1
• Filed: 06/30/2004
• Published: 01/05/2006
• Est. Priority Date: 06/30/2004
• Status: Active Grant

• Current Assignee: Marvell Asia Pte Limited (Marvell Technology Group Limited)
• Original Assignee: Marvell International Limited (Marvell Technology Group Limited)
• Inventors: Erell, Adoram, Aronowitz, Hagai
• Granted Patent: US 7,469,205 B2
• Time in Patent Office: Days
• US Class Current: 704/10
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G10L 13/08 Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination
• CPC Class Codes: G10L 15/06 Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice G10L15/14 takes precedence

Abstract

A compressed pronunciation lexicon file is generated from a source pronunciation lexicon using a pronunciation prediction algorithm in a multi-output mode. The pronunciation prediction algorithm may generate a deterministic ordered list of phoneme strings from the textual representation of a particular word. The compressed pronunciation lexicon file may include a sorted list of records of compressed textual representations of words and compressed phonetic representations of the words.

Claims

1. A method comprising:

generating a compressed pronunciation lexicon file from a source pronunciation lexicon using a pronunciation prediction algorithm in a multi-output mode.

2. The method of claim 1, wherein using the pronunciation prediction algorithm includes using a pronunciation prediction algorithm that is based on a hidden Markov model.

3. The method of claim 1, wherein the source lexicon includes a list of textual representations of words and corresponding phonetic representations of the words, and wherein using the pronunciation prediction algorithm in the multi-output mode includes:

generating a deterministic ordered list of phoneme strings for a particular one of the words.

4. The method of claim 3, wherein a particular one of the phoneme strings is identical to the corresponding phonetic representation of the particular word, and generating the compressed pronunciation lexicon file further includes:

substituting the corresponding phonetic representation of the particular word with an index of the particular one of the phoneme strings.

5. The method of claim 3, wherein none of the phoneme strings is identical to corresponding phonetic representation of the particular word, and generating the compressed pronunciation lexicon file further includes:

substituting the corresponding phonetic representation of the particular word with an index of a most closely matching one of the phoneme strings and an encoded representation of edit operations to convert the most closely matching phoneme string into the corresponding phonetic representation of the particular word.

6. The method of claim 1, wherein the source lexicon includes a list of textual representations of words and corresponding phonetic representations of the words, and wherein generating the compressed lexicon file further includes:

sorting the list in alphabetical order of the textual representations to generate a sorted list; and

substituting a textual representation of a particular word with i) the length of a common string of initial letters of the textual representation and the textual representation of a preceding word in the list, and ii) an encoded version of the remaining letters of the textual representation of the particular word.

7. The method of claim 6, wherein generating the compressed lexicon file further includes:

generating a Huffman-encoded version of the remaining letters of the textual representation of the particular word and embedding a Huffman coding table or tree in the compressed file.

8. The method of claim 6, further comprising:

embedding an index table in the compressed file, the index table containing pointers to selected records of the sorted list such that the selected records are substantially evenly distributed along the sorted list.

9. A method for extracting a pronunciation of a word from a compressed pronunciation lexicon file, the method comprising:

using a pronunciation prediction algorithm in a multi-output mode to generate a deterministic ordered list of phoneme strings from a textual representation of the word;

if a record in the compressed pronunciation lexicon file corresponding to the word includes an index of a particular phoneme string from the deterministic ordered list and an indication of edit operations, editing the particular phoneme string according to the edit operations to produce the pronunciation; and

otherwise, if the record includes the index, identifying the particular phoneme string as the pronunciation.

10. The method of claim 9, wherein using the pronunciation prediction algorithm includes using a pronunciation prediction algorithm that is based on a hidden Markov model.

11. The method of claim 9, further comprising:

identifying the record in the compressed pronunciation lexicon file that corresponds to the word by serially decompressing entries of the record and preceding records.

12. The method of claim 11, wherein serially decompressing the entries begins at a record pointed to by a pointer in an index table of the compressed pronunciation lexicon file.

13. An article comprising a computer-readable storage medium having stored thereon instructions that, when executed by a processor, result in:

generating a compressed pronunciation lexicon file from a source pronunciation lexicon using a pronunciation prediction algorithm in a multi-output mode.

14. The article of claim 13, wherein the pronunciation prediction algorithm is based on a hidden Markov model.

15. The article of claim 13, wherein the instructions further result in:

generating a deterministic ordered list of phoneme strings for a particular one of the words.

16. The article of claim 15, wherein if a particular one of the phoneme strings is identical to the corresponding phonetic representation of the particular word, the instructions further result in:

substituting the corresponding phonetic representation of the particular word with an index of the particular one of the phoneme strings.

17. The article of claim 15, wherein if none of the phoneme strings is identical to corresponding phonetic representation of the particular word, the instructions further result in:

substituting the corresponding phonetic representation of the particular word with an index of a most closely matching one of the phoneme strings and an encoded representation of edit operations to convert the most closely matching phoneme string into the corresponding phonetic representation of the particular word.

18. An article comprising a computer-readable storage medium having stored thereon a compressed pronunciation lexicon file comprising:

a sorted list of records of textual representations of words and compressed phonetic representations of the words, the compressed phonetic representations generated using a pronunciation prediction algorithm in a multi-output mode.

19. The article of claim 18, wherein the textual representations of words are compressed textual representations including prefix lengths and encoded suffixes, wherein a prefix length of a particular word is the length of a common string of initial letters of a textual representation of the particular word and the textual representation of a preceding word in the sorted list, and wherein an encoded suffix is an encoded version of the remaining letters of the textual representation of the particular word.

20. The article of claim 19, wherein the compressed pronunciation lexicon further comprises:

an index table including pointers to selected records of the sorted list such that the selected records are substantially evenly distributed along the sorted list.

21. An apparatus comprising:

a memory to store a compressed pronunciation lexicon file and a pronunciation prediction module that implements a pronunciation prediction algorithm; and

a processor to extract pronunciations from the compressed pronunciation lexicon file using the pronunciation prediction algorithm in a multi-output mode.

22. The apparatus of claim 21, wherein the pronunciation prediction algorithm is based on a hidden Markov model.

23. The apparatus of claim 21, wherein the compressed pronunciation lexicon file includes a sorted list of records of textual representations of words and compressed phonetic representations of the words, the compressed phonetic representations generated using the pronunciation prediction algorithm in the multi-output mode.

24. An apparatus comprising:

an antenna;

a memory to store a compressed pronunciation lexicon file and a pronunciation prediction module that implements a pronunciation prediction algorithm; and

a processor to extract pronunciations from the compressed pronunciation lexicon file using the pronunciation prediction algorithm in a multi-output mode.

25. The apparatus of claim 24, wherein the pronunciation prediction algorithm is based on a hidden Markov model.

26. The apparatus of claim 24, wherein the compressed pronunciation lexicon file includes a sorted list of records of textual representations of words and compressed phonetic representations of the words, the compressed phonetic representations generated using the pronunciation prediction algorithm in the multi-output mode.