ENHANCED HUMAN MACHINE INTERFACE THROUGH HYBRID WORD RECOGNITION AND DYNAMIC SPEECH SYNTHESIS TUNING

• US 20150206539A1
• Filed: 06/04/2014
• Published: 07/23/2015
• Est. Priority Date: 06/04/2013
• Status: Abandoned Application

• Current Assignee: Ridetones, Inc.
• Original Assignee: IMS Solution, Inc.
• Inventors: Campbell, David Neil, El-Ghazal, Akrem Saad, Rae, Robert Andrew, Sulpizi, Daniel John Vincent
• Application Number:
  US14/296,044
• Publication Number:
  US 20150206539A1
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G10L 13/02 Methods for producing synthetic speech; Speech synthesisers
• 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 15/08 Speech classification or search
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 17/22 Interactive procedures; Man-machine interfaces

Abstract

A human machine interface enables human users to interact with a machine by inputting auditory and/or textual data. The interface and corresponding method perform efficient look up of words, corresponding to inputted human data, which are stored in a domain database. The robustness of a speech synthesis engine is enhanced by updating the deployed pronunciation vocabulary dynamically. The architecture of the preferred embodiment of the former method includes a combination of ensemble matching, clustering, and rearrangement methods. The latter method involves retrieving suggested phonetic pronunciations for words unknown to the speech synthesis engine and verifying those through a manual or autonomous process.

Claims

1. A method to perform word look-up based on human input including the steps of:

receiving human input;

performing initial recognition of the human input;

receiving metadata based upon the initial recognition;

prioritizing a plurality of possible words based upon the metadata; and

outputting a first word of the plurality of possible words based upon the prioritization.

2. The method in claim 1 wherein the human input is voice-based.

3. The method in claim 1 wherein the human input is text-based.

4. The method of claim 1 wherein the plurality of possible words and their associated metadata are stored in a domain database.

5. The method of claim 1 wherein recognition of human input voice data is performed using a voice recognizer to produce a set of potential words.

6. The method of claim 5 wherein a human input recognizer may be running locally or on a remote server residing in a cloud.

7. The method of claim 1 wherein a set of potential words are matched against the plurality of possible words using an ensemble of matching methods.

8. The method of claim 7 wherein an ensemble of nearest-neighbor methods operates by minimizing a weighted aggregate of potential to possible word distances.

9. The method of claim 8 wherein the potential to possible word distances are computed in two or more spaces.

10. The method of claim 9 where one space is phonetic encoding.

11. The method of claim 9 where one space is double metaphone encodings.

12. The method of claim 9 where one space is a natural edit distance.

13. The method of claim 1 wherein a set of candidate words is obtained by sorting a set of possible words according to their computed aggregate distance and outputting only a predefined number of top words.

14. The method of claim 1 wherein a set of produced candidate words are processed into two clusters of distinct words, and relevant/irrelevant words.

15. The method of claim 14 wherein the clustering is performed using a segmentation method.

16. The method of claim 14 wherein a set of produced distinct words can be rearranged according to their corresponding metadata of interest to produce a set of recognized words.

17. The method of claim 1 wherein the metadata includes frequency of usage.

18. A method of performing text to speech processing:

receiving text input including a plurality of words;

performing word recognition on the text input;

identifying native words of the plurality of words that already existing in a phonetic vocabulary; and

identifying alien words of the plurality of words that do not exist in the phonetic vocabulary.

19. The method of claim 18 wherein a vocabulary of words and their corresponding verified pronunciation are stored in the phonetic vocabulary.

20. The method of claim 19 further including the steps of dynamically retrieving through a remote server inquiry a suggested phonetic pronunciation for the alien word.

21. The method of claim 20 further including the step of validating a suggested phonetic pronunciation for the alien word by a human agent.

22. The method of claim 21 further including the step of adding the suggested phonetic pronunciation to the phonetic vocabulary based upon being validated by the human agent.

23. The method of claim 20 further including the step of validating a suggested phonetic pronunciation by a software agent.

24. The method of claim 21 further including the step of adding the suggested phonetic pronunciation to the phonetic vocabulary based upon being validated by the software agent.