FLEXIBLE ARCHITECTURE FOR ACOUSTIC SIGNAL PROCESSING ENGINE

• US 20140188470A1
• Filed: 12/31/2012
• Published: 07/03/2014
• Est. Priority Date: 12/31/2012
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Chang, Jenny, Deisher, Michael E., Iyer, Ravishankar
• Granted Patent: US 9,653,070 B2
• Time in Patent Office: Days
• US Class Current: 704/236
• CPC Class Codes: G10L 15/14 using statistical models, e.g. Hidden Markov Models [HMMs] G10L15/18 takes precedence

Abstract

A disclosed speech processor includes a front end to receive a speech input and generate a feature vector indicative of a portion of the speech input and a Gaussian mixture (GMM) circuit to receive the feature vector, model any one of a plurality of GMM speech recognition algorithms, and generate a GMM score for the feature vector based on the GMM speech recognition algorithm modeled. In at least one embodiment, the GMM circuit includes a common compute block to generate feature a vector sum indicative of a weighted sum of differences squares between the feature vector and a mixture component of the GMM speech recognition algorithm. In at least one embodiment, the GMM speech recognition algorithm being modeled includes a plurality of Gaussian mixture components and the common compute block is operable to generate feature vector scores corresponding to each of the plurality of mixture components.

Claims

8. A device, comprising:

a processing core to execute instruction set instructions;

storage to store instructions executable by the processing core;

a Gaussian mixture (GMM) score generator to receive algorithm inputs indicative of any of a plurality of GMM speech recognition algorithms and generate a GMM score for the feature vector based on the GMM speech recognition algorithm indicated by the algorithm inputs;

a microphone; and

a microcontroller to;

group audio samples from the microphone into blocks and perform feature extraction on the group;

invoke the GMM score generator to score extracted features;

indicate, based on the generated scores, whether the audio was speech; and

wake up features of the device responsive to determining that the audio was speech.

9. The device of claim 8, further comprising.

a power manager to receive power from a battery and provide to the processor;

a radio frequency transceiver to establish a wireless communication transport;

a touchscreen display to receive touch input;

an audio coder/decoder to receive audio input and generate a feature vector comprising a digital representation the audio input; and

10. The device of claim 9, wherein the instructions include instructions executable by the processing core to receive the GMM score and initiate a search for text based on the GMM score and wherein the instructions executable to initiate the search include instructions executable to transmit the GMM score via the wireless communication transport to a remote search engine.

11. The device of claim 8, wherein the GMM scoring logic includes a sum of differences squared circuit to receive the feature vector and generate a GMM sum indicative of a weighted sum of differences squared between the feature vector and a mixture component of the indicated GMM speech recognition algorithm.

12. The device of claim 11, wherein the instructions include speech recognition application instructions to invoke the sum of differences squared circuit for each of a plurality of mixture components associated with the GMM speech recognition algorithm indicated to generate a plurality of GMM sums corresponding to the feature vector.

13. The device of claim 12, wherein the GMM scoring logic includes score processing logic to determine the GMM score based on the plurality of GMM sums received.

14. The device of claim 13, wherein the score processing logic is operable to receive configuration inputs indicative of a score processing option selected from a logarithmic summation option and a maximum sum option.

15. A processor, comprising:

a processing core to execute instruction set instructions;

an audio interface to receive feature vector data comprising a digital representation of a speech sample; and

a Gaussian mixture (GMM) score generator to generate a GMM score corresponding to the feature vector, wherein the GMM scoring logic includes;

weighted sum of differences squared (SODS) logic to compute a GMM sum indicative of differences between elements of the feature vector and corresponding elements of a GMM component mixture.

16. The processor of claim 15, wherein the GMM scoring logic includes algorithm selection logic to select a first GMM scoring algorithm from a plurality of supported GMM scoring algorithms.

17. The processor of claim 15, wherein the plurality of supported GMM scoring algorithms include a logarithmic summation scoring algorithm.

18. The processor of claim 15, wherein the plurality of supported GMM scoring algorithms include a maximum summation scoring algorithm.

19. The processor of claim 15, wherein the SODS logic includes a plurality of stages, each of the plurality of stages including a plurality of circuits to determine:

a squared value of a difference between two inputs from a preceding stage; and

a weighting to apply to the squared value.

20. A computer readable storage medium including processor executable instructions, which when executed by the processor, cause the processor to:

provide algorithm selection inputs indicative of a first speech recognition algorithm to Gaussian mixture (GMM) scoring logic;

detect a feature vector comprising a digital representation of an interval of speech;

invoke the GMM scoring logic to generate a GMM score for the feature vector based on the first speech recognition algorithm; and

initiate a search for text based on the GMM score.

1. A speech processor, comprising:

an input to receive a feature vector indicative of a portion of the speech input; and

Gaussian mixture (GMM) score generator logic to receive the feature vector, model any of a plurality of GMM speech recognition algorithms, and generate a GMM score for the feature vector based on the GMM speech recognition algorithm modeled.

2. The speech processor of claim 1, wherein the portion of speech corresponds to an interval of the speech input having a specified duration.

3. The speech processor of claim 1, wherein the GMM circuit includes a common compute block to generate feature a vector sum indicative of a weighted sum of differences squared between the feature vector and a mixture component of the GMM speech recognition algorithm.

4. The speech processor of claim 3, wherein the mixture component includes a mean vector and a variance vector and wherein the feature vector score is indicative of squared differences between values of the feature vector and corresponding values of the mean vector, weighted by corresponding value of the variance vector.

5. The speech processor of claim 1, wherein the GMM speech recognition algorithm includes a plurality of mixture components and the common compute block is operable to generate feature vector score corresponding to each of the plurality of mixture components.

6. The speech processor of claim 5, wherein the GMM circuit includes a score selection block to:

receive the feature vector scores and algorithm inputs;

implement the modeled GMM speech recognition algorithm based on the algorithm inputs; and

generate the GMM score for the modeled speech recognition algorithm based on the feature vector scores.

7. The speech processor of claim 1, wherein the algorithm inputs include a logarithmic mode input indicative of whether the modeled GMM speech recognition algorithm implements a logarithmic summation calculation to determine the GMM score.

21. The computer readable storage medium of claim 20, wherein the instructions to initiate the search include instructions to transmit the GMM score wirelessly to a remotely located speech recognition database.

22. The computer readable storage medium of claim 20, wherein the instructions include instructions to receive a wirelessly transmit search result including search result text.

23. The computer readable storage medium of claim 20, wherein the feature vector represents an interval of speech received from an audio input.

22-1. The computer readable storage medium of claim 20, wherein the instructions further include instructions to display the text on a display device.

24. A tangible machine readable medium having stored thereon a set of information representing hardware logic circuits, which if used by a machine, causes the machine to fabricate hardware logic circuits, including:

a first logic circuit to receive a feature vector and compute a sum of weighted differences squared value based on the feature vector and a mixture vector including a mean vector and a variance vector; and

a second logic circuit to receive, from the first logic, the sum of weighted differences squared values for each of a plurality of mixture vectors associated with a Gaussian mixture mode implementation and generate a Gaussian mixture mode score based on a Gaussian mixture mode scoring algorithm.

25. The tangible machine readable medium of claim 24, wherein the second logic circuit is operable to receive values for algorithm parameters wherein the algorithm parameter values determine the Gaussian mixture mode algorithm from a set of supported scoring algorithms.

26. The tangible machine readable medium of claim 25, wherein the set of supported scoring algorithms include an algorithm that employs a logarithmic summation of the sum of weighted differences squared values.

27. The tangible machine readable medium of claim 24, further comprising:

a processing core logic circuit to execute instruction set instructions;

a power management logic circuit;

a platform controller hub circuits; and

touchscreen display adapter circuit.

28. The tangible machine readable medium of claim 27, further comprising:

a radio frequency circuit to establish wireless communication.