Detection and use of acoustic signal quality indicators

• US 8,812,326 B2
• Filed: 08/06/2013
• Issued: 08/19/2014
• Est. Priority Date: 04/03/2006
• Status: Active Grant

• Current Assignee: Promptu Systems Corporation
• Original Assignee: Promptu Systems Corporation
• Inventors: Chittar, Naren, Gulati, Vikas, Pratt, Matthew, Printz, Harry
• Primary Examiner(s): SAINT CYR, LEONARD
• Application Number:
  US13/960,714
• Publication Number:
  US 20130325485A1
• Time in Patent Office: 378 Days
• Field of Search: 704/246, 704/247, 704/251, 704/252, 704/275
• US Class Current: 704/275
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 21/06 Transformation of speech into a non-audible representation, e.g. speech visualisation or speech processing for tactile aids G10L15/26 takes precedence

Abstract

A computer-driven device assists a user in self-regulating speech control of the device. The device processes an input signal representing human speech to compute acoustic signal quality indicators indicating conditions likely to be problematic to speech recognition, and advises the user of those conditions.

Claims

1. A computer implemented method to regulate a speaker'"'"'s issuance of speech based commands, the method comprising operations of:

receiving an input signal representing audio content including speech of the speaker at a microphone, the audio content occurring between a prescribed begin event and a prescribed end event;

processing at least a portion of the input signal to determine whether the processed portion of the input signal contains speech exhibiting one or more predetermined conditions capable of posing difficulty in successfully performing speech recognition upon the input signal;

where the processed portion of the input signal comprises one of;

all of the input signal, a part of the input signal, speech present in the input signal;

if the processing operation determines that the processed portion of the input signal contains speech exhibiting one or more of the predetermined conditions, then performing operations comprising;

issuing at least one human comprehensible speech quality alert corresponding to each exhibited predetermined condition;

the processing operation comprising determining whether the microphone is too near or too far on the basis of a near field effect, wherein low-frequency responsiveness of the microphone to the speaker varies with distance between the speaker and the microphone;

wherein as separation of the speaker and the microphone increases, the amount of low-frequency acoustic energy captured by the microphone decreases, with respect to the total amount of acoustic energy captured from the speaker.

2. The method of claim 1, wherein computation of a microphone too near condition proceeds by comparing the amount of low-frequency acoustic energy in the speaker'"'"'s utterance with the amount of acoustic energy in all frequencies in the utterance;

wherein if said comparison shows an excessively large amount of low-frequency acoustic energy, the microphone is too near.

3. The method of claim 2, wherein the determination is based upon the accumulation, by frequency, of acoustic energy in the utterance.

4. The method of claim 2, wherein the determination is based upon estimating the slope of a curve fitted to the low-frequency portion of the acoustic energy spectrum of the utterance.

5. The method of claim 3, wherein acoustic energy present in the low-frequency portion of the utterance is accumulated, along with the total acoustic energy present in the utterance;

wherein a ratio of these two energy quantities is compared with a prespecified fixed value; and

wherein if the ratio exceeds said fixed value, the microphone is too near.

6. The method of claim 4, wherein acoustic energy present in each of a plurality of low-frequency fast Fourier transform (FFT) bins is accumulated over the utterance, for each bin;

wherein a resulting set of <

bin number, energy in bin>

pairs is treated as a collection of samples of a continuous function, which represents the estimated acoustic energy spectrum of the utterance;

wherein a least-mean-squares linear fit to said function is computed, and a slope of a resulting linear equation is extracted;

wherein a value of said slope is compared to a predetermined fixed value;

wherein if said slope is too low, the microphone is too near.

7. The method of claim 1, wherein computation of a microphone too far condition proceeds by comparing an amount of low-frequency acoustic energy in the utterance with an amount of acoustic energy in all frequencies in the utterance;

wherein if this comparison shows an excessively small amount of low-frequency acoustic energy the microphone is too far.

8. The method of claim 7, wherein the determination is based upon the accumulation, by frequency, of acoustic energy in the utterance.

9. The method of claim 7, wherein the determination is based upon estimating the slope of a curve fitted to the low-frequency portion the acoustic energy spectrum of the utterance.

10. The method of claim 8, wherein acoustic energy present in the low-frequency portion of the utterance is accumulated, along with the total acoustic energy present in the utterance;

wherein a ratio of these two quantities is compared with a prespecified fixed value;

wherein if the ratio falls below the fixed value the microphone is too far.

11. The method of claim 9, wherein acoustic energy present in each of a plurality of low-frequency fast Fourier transform (FFT) bins is accumulated over the utterance for each bin;

wherein a resulting set of <

bin number, energy in bin>

pairs is treated as a collection of samples of a continuous function, which represents the estimated acoustic energy spectrum of the utterance;

wherein a least-mean-squares linear fit to this function is computed, and the slope of the resulting linear equation is extracted;

wherein a value of said slope is compared to a predetermined fixed value;

wherein if said slope is too high, the microphone is too far.