Speech signal similarity

• US 8,670,983 B2
• Filed: 08/30/2011
• Issued: 03/11/2014
• Est. Priority Date: 09/02/2010
• Status: Active Grant

• Current Assignee: Nexidia Incorporated (Nice Ltd)
• Original Assignee: Nexidia Incorporated (Nice Ltd)
• Inventors: Gavalda, Marsal, Garland, Jacob B., Arrowood, Jon A., Lanham, Drew
• Primary Examiner(s): Vo, Huyen X.
• Application Number:
  US13/221,270
• Publication Number:
  US 20120059656A1
• Time in Patent Office: 924 Days
• Field of Search: 704/231, 704/251, 704/254, 704/235, 704/240, 704/243, 704/244, 704/245, 704/249, 704/250
• US Class Current: 704/254
• CPC Class Codes: G10L 25/00 Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 muting semiconductor-based amplifiers when some special characteristics of a signal are sensed by a speech detector, e.g. sensing when no signal is present, H03G3/34

Abstract

A method for determining a similarity between a first audio source and a second audio source includes: for the first audio source, determining a first frequency of occurrence for each of a plurality of phoneme sequences and determining a first weighted frequency for each of the plurality of phoneme sequences based on the first frequency of occurrence for the phoneme sequence; for the second audio source, determining a second frequency of occurrence for each of a plurality of phoneme sequences and determining a second weighted frequency for each of the plurality of phoneme sequences based on the second frequency of occurrence for the phoneme sequence; comparing the first weighted frequency for each phoneme sequence with the second weighted frequency for the corresponding phoneme sequence; and generating a similarity score representative of a similarity between the first audio source and the second audio source based on the results of the comparing.

Claims

1. A method for determining a similarity between a first audio source and a second audio source, the method comprising:

for the first audio source, performing the steps of;

determining, using an analysis module of a computer, a first plurality of segments of the first audio source;

determining, using the analysis module, a first frequency of occurrence for each of a plurality of phoneme sequences in the first audio source;

determining, using the analysis module, a first weighted frequency for each of the plurality of phoneme sequences based on the first frequency of occurrence for the phoneme sequence;

wherein determining the first weighted frequency includes emphasizing phoneme sequences that occur in few segments of the first plurality of segments relative to phoneme sequences that occur in many segments of the first plurality of segments;

for the second audio source, performing the steps of;

determining, using the analysis module, a second plurality of segments of the second audio source;

determining, using the analysis module, a second frequency of occurrence for each of a plurality of phoneme sequences in the second audio source;

determining, using the analysis module, a second weighted frequency for each of the plurality of phoneme sequences based on the second frequency of occurrence for the phoneme sequence;

wherein determining the second weighted frequency includes emphasizing phoneme sequences that occur in few segments of the second plurality of segments relative to phoneme sequences that occur in many segments of the second plurality of segments;

comparing, using a comparison module of a computer, the first weighted frequency for each phoneme sequence with the second weighted frequency for the corresponding phoneme sequence; and

generating, using the comparison module, a similarity score representative of a similarity between the first audio source and the second audio source based on the results of the comparing.

2. The method of claim 1, wherein determining the first frequency of occurrence includes, for each phoneme sequence, determining a ratio between a number of times the phoneme sequence occurs in the first audio source and a duration of the first audio source.

3. The method of claim 1, wherein the first weighted frequencies for each first portion of audio are collectively represented by a first vector and the second weighted frequencies for each second portion of audio are collectively represented by a second vector.

4. The method of claim 3, wherein the step of comparing includes determining a cosine of an angle between the first vector and the second vector.

5. The method of claim 1, wherein the step of comparing includes using a latent semantic analysis technique.

6. The method of claim 1, wherein the first audio source forms a part of a first audio file and the second audio source forms a part of a second audio file.

7. The method of claim 1, wherein the first audio source is a first segment of an audio file and the second audio source is a second segment of the audio file.

8. The method of claim 1, further comprising selecting the plurality of phoneme sequences.

9. The method of claim 8, wherein the plurality of phoneme sequences are selected on the basis of a language of at least one of the first audio source and the second audio source.

10. The method of claim 1, wherein each phoneme sequence includes three phonemes.

11. The method of claim 1, wherein each phoneme sequence includes a plurality of words.

12. The method of claim 11, further comprising determining a relevance score for each word in the first audio source.

13. The method of claim 12, wherein the relevance score for each word is determined based on a frequency of occurrence of the word in the first audio source.

14. A method for determining a similarity between a first audio source and a second audio source, the method comprising:

generating, using a computer, a phonetic transcript of the first audio source, the phonetic transcript including a list of phonemes occurring in the first audio source;

selecting a plurality of sequences of phonemes from the list of phonemes, each sequence of phonemes being associated with a time interval in the first audio source;

searching, using the computer, the second audio source to identify occurrences of each of the plurality of sequences of phonemes, each identified occurrence being associated with a time interval in the second audio source and a search score;

forming a set of merged sequences of phonemes including merging at least some sequences of phonemes of the plurality of sequences of phonemes with overlapping time intervals;

forming a set of merged occurrences of sequences of phonemes including merging occurrences of sequences of phonemes with overlapping time intervals, including for each merged occurrence, forming an associated score by accumulating the search scores associated with the occurrences and forming an associated time duration by accumulating time durations associated with the occurrences;

and

generating, using the computer, a score representative of a similarity between the first audio source and the second audio source, based on one or both of;

the scores associated with the merged set of occurrences of sequences of phonemes and the time durations associated with the merged set of occurrences of sequences of phonemes.

15. The method of claim 14, wherein the phonetic transcript includes a sequential list of phonemes occurring in the first audio source.