Automatic speech recognition method and apparatus

• US 8,311,825 B2
• Filed: 10/03/2008
• Issued: 11/13/2012
• Est. Priority Date: 10/04/2007
• Status: Active Grant

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation), Toshiba Digital Solutions Corporation (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Chen, Langzhou
• Primary Examiner(s): Chawan, Vijay B
• Application Number:
  US12/244,997
• Publication Number:
  US 20090099841A1
• Time in Patent Office: 1,502 Days
• Field of Search: 704/9, 704/240, 704/255, 704/254, 704/246, 704/251, 704/243, 704/256, 704/256.2, 704/257
• US Class Current: 704/240
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/262 using context analysis, e.g. lexical, syntactic or semantic context
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams

Abstract

A system for calculating the look ahead probabilities at the nodes in a language model look ahead tree, wherein the words of the vocabulary of the language are located at the leaves of the tree,

• • said apparatus comprising:
  • means to assign a language model probability to each of the words of the vocabulary using a first low order language model;
  • means to calculate the language look ahead probabilities for all nodes in said tree using said first language model;
  • means to determine if the language model probability of one or more words of said vocabulary can be calculated using a higher order language model and updating said words with the higher order language model; and
  • means to update the look ahead probability at only the nodes which are affected by the words where the language model has been updated.

Claims

1. A system for calculating the look ahead probabilities at the nodes in a language model look ahead tree, wherein the words of the vocabulary of the language are located at the leaves of the tree,

said system comprising a processor adapted to:

to obtain a first low order language model look ahead tree having a language first low order language model probability assigned to the words of the vocabulary and the language model look ahead probabilities calculated for nodes in said first low order language model look ahead tree;

to determine if the language model probability of one or more words of said vocabulary can be calculated using a higher order language model and updating said words with the higher order language model; and

to update the look ahead probability at the nodes of the first low order language model look ahead tree which are affected by the words where the language model has been updated, wherein the first low order language model look ahead tree is one order lower than the higher order language model.

2. A system according to claim 1, wherein said processor which is adapted to assign a language model probability using the lower order language model is configured to use a back-off parameter.

3. A system according to claim 1, wherein said processor which is adapted to determine if the language model probability of one or more words of said vocabulary can be calculated using a higher order language model is configured to determine if a word has occurred in combination with a previously recognised word.

4. A system according to claim 1, wherein the higher order model is a bigram, trigram, fourgram or higher order n-gram model.

5. A system according to claim 4, further comprising a memory having a multi-cache structure, wherein the number of caches of the memory equals the order of the n-gram model used in the system.

6. An automatic speech recognition system comprising:

a receiver adapted to receive speech;

a processor adapted to,

extract acoustic signals from said speech and determine a first probability of the words contained in said speech,

provide a language model to determine a second probability of the words contained in said speech, comprising a system for calculating the look ahead probabilities at the nodes in a language model look ahead tree in accordance with claim 1, and

to combine the first and second probabilities and output the words recognised from the received speech.

7. A speech to speech translation system comprising the automatic speech recognition system of claim 6 and a processor adapted to translate the outputted words into a further language.

8. An OCR system comprising:

a receiver adapted to receive scanned text; and

a processor adapted to,

determine a first probability of the words contained in said text,

provide a language model to determine a second probability of the words contained in said text, comprising a system for calculating the look ahead probabilities at the nodes in a language model look ahead tree in accordance with claim 1, and

combine the first and second probabilities and output the words recognised from the scanned text in a format which can be read by a computer.

9. A handwriting recognition system comprising:

a receiver adapted to receive handwritten text; and

a processor adapted to,

determine a first probability of the words contained in said text,

provide a language model to determine a second probability of the words contained in said text, comprising a system for calculating the look ahead probabilities at the nodes in a language model look ahead tree in accordance with claim 1, and

combine the first and second probabilities and output the words recognised from the handwritten text.

10. A system according to claim 1, wherein said processor is adapted to update one or more words with a higher order language model probability a plurality of times in an iterative manner and increase the order of language model look ahead tree by 1 for each iteration.

11. A system according to claim 1, wherein the lower order language model look ahead tree is a language model look ahead tree which has been previously updated, the order of said updated lower order language model look ahead tree being the order of the highest order language model used to update the updated lower order-language model look ahead tree.

12. An automatic speech recognition system according to claim 11, further comprising a cache and wherein said processor is adapted to retrieve said previously updated lower order language model look ahead tree from said cache.

13. A system according to claim 1, wherein the first low order language model look ahead tree is an n−

1 gram language model look ahead tree and said higher order language model is an n-gram language model, and n is an integer of at least 2.

14. A system according to claim 10, wherein in the first iteration the language model look ahead probabilities are updated using a bigram language model and then in the second iteration, the updated model is further updated using a trigram language model probabilities.

15. A system according to claim 1, wherein said processor is adapted to cache said lower order language model look ahead tree.

16. A method of calculating language model look ahead probabilities at the nodes in a language model look ahead tree, wherein the words of the vocabulary of the language are located at the leaves of the tree,

said method comprising:

assigning a language model probability to each of the words of the vocabulary using a first low order language model;

calculating the language look ahead probabilities for all nodes in said tree using said first language model;

determining if the language model probability of one or more words of said vocabulary can be calculated using a higher order language model and updating said words with the higher order language model probabilities; and

updating the look ahead probability at only the nodes which are affected by the words where the language model has been updated.

17. A method according to claim 16, wherein determining the lower order language model probabilities uses a back-off parameter.

18. A method according to claim 16, wherein determining if the language model probability of one or more words of said vocabulary can be calculated using a higher order language model determines if a word has occurred in combination with a previously recognised word.

19. A method according to claim 16, wherein the higher order model is a bigram, trigram, fourgram or higher order n-gram model.

20. A method according to claim 16, further comprising:

decoding a subsequent signal by resetting the language model look ahead probabilities to the lower order language model;

determining if the language model probability of one or more words of said vocabulary can be calculated using a higher order language model and updating said words with the higher order language model probabilities; and

updating the look ahead probability at only the nodes which are affected by the words where the language model has been updated.

21. An automatic speech recognition method comprising:

receiving speech;

extracting acoustic signals from said speech and determining a probability of the words contained in said speech based on said acoustic signals;

providing a language model, comprising calculating the look ahead probabilities at the nodes in a language model look ahead tree in accordance with claim 16; and

outputting the words recognised from the received speech.

22. A speech to speech translation method comprising the automatic speech recognition method of claim 21 and translating of the outputted words into a further language.

23. An OCR method comprising:

receiving scanned text;

determining a first probability of the words contained in said text;

providing a language model to determine a second probability of the words contained in said text, comprising a method for calculating the look ahead probabilities at the nodes in a language model look ahead tree in accordance with claim 16; and

combining the first and second probabilities and output the words recognised from the scanned text in a format which can be read by a computer.

24. A handwriting recognition method comprising:

receiving handwritten text;

determining a first probability of the words contained in said text;

providing a language model to determine a second probability of the words contained in said text, comprising a method for calculating the look ahead probabilities at the nodes in a language model look ahead tree in accordance with claim 16; and

combining the first and second probabilities and output the words recognised from the handwritten text.

25. A non-transitory computer readable medium storing program instructions for controlling a computer to carry out the method of claim 16.