Patents by Inventor Arvindh Krishnaswamy

Arvindh Krishnaswamy has filed for patents to protect the following inventions. This listing includes patent applications that are pending as well as patents that have already been granted by the United States Patent and Trademark Office (USPTO).

  • Publication number: 20190251974
    Abstract: A speech recognition system for resolving impaired utterances can have a speech recognition engine configured to receive a plurality of representations of an utterance and concurrently to determine a plurality of highest-likelihood transcription candidates corresponding to each respective representation of the utterance. The recognition system can also have a selector configured to determine a most-likely accurate transcription from among the transcription candidates. As but one example, the plurality of representations of the utterance can be acquired by a microphone array, and beamforming techniques can generate independent streams of the utterance across various look directions using output from the microphone array.
    Type: Application
    Filed: April 19, 2019
    Publication date: August 15, 2019
    Inventors: Sean A. Ramprashad, Harvey D. Thornburg, Arvindh Krishnaswamy, Aram M. Lindahl
  • Patent number: 10304462
    Abstract: A speech recognition system for resolving impaired utterances can have a speech recognition engine configured to receive a plurality of representations of an utterance and concurrently to determine a plurality of highest-likelihood transcription candidates corresponding to each respective representation of the utterance. The recognition system can also have a selector configured to determine a most-likely accurate transcription from among the transcription candidates. As but one example, the plurality of representations of the utterance can be acquired by a microphone array, and beamforming techniques can generate independent streams of the utterance across various look directions using output from the microphone array.
    Type: Grant
    Filed: January 15, 2018
    Date of Patent: May 28, 2019
    Assignee: Apple Inc.
    Inventors: Sean A. Ramprashad, Harvey D. Thornburg, Arvindh Krishnaswamy, Aram M. Lindahl
  • Patent number: 10013981
    Abstract: A speech recognition system for resolving impaired utterances can have a speech recognition engine configured to receive a plurality of representations of an utterance and concurrently to determine a plurality of highest-likelihood transcription candidates corresponding to each respective representation of the utterance. The recognition system can also have a selector configured to determine a most-likely accurate transcription from among the transcription candidates. As but one example, the plurality of representations of the utterance can be acquired by a microphone array, and beamforming techniques can generate independent streams of the utterance across various look directions using output from the microphone array.
    Type: Grant
    Filed: June 6, 2015
    Date of Patent: July 3, 2018
    Inventors: Sean A. Ramprashad, Harvey D. Thornburg, Arvindh Krishnaswamy, Aram M. Lindahl
  • Publication number: 20180137864
    Abstract: A speech recognition system for resolving impaired utterances can have a speech recognition engine configured to receive a plurality of representations of an utterance and concurrently to determine a plurality of highest-likelihood transcription candidates corresponding to each respective representation of the utterance. The recognition system can also have a selector configured to determine a most-likely accurate transcription from among the transcription candidates. As but one example, the plurality of representations of the utterance can be acquired by a microphone array, and beamforming techniques can generate independent streams of the utterance across various look directions using output from the microphone array.
    Type: Application
    Filed: January 15, 2018
    Publication date: May 17, 2018
    Inventors: Sean A. Ramprashad, Harvey D. Thornburg, Arvindh Krishnaswamy, Aram M. Lindahl
  • Patent number: 9917562
    Abstract: Automatic gain control systems disclosed herein can incorporate a confidence metric that can estimate the accuracy of gain adjustments calculated by an automatic gain control module. The confidence metric may be based on a percentage of valid audio samples in a given period of time. Based on the confidence metric, the AGC response may be reduced, delayed, frozen, or otherwise altered from the baseline gain adjustment. Time-averaging process may be used to estimate the input signal power level and determine an appropriate baseline gain adjustment. Additionally, weighting functions can be adjusted to prevent overestimation of the signal power.
    Type: Grant
    Filed: June 7, 2016
    Date of Patent: March 13, 2018
    Assignee: Apple Inc.
    Inventors: Arvindh Krishnaswamy, Juha O. Merimaa, Kapil Krishnamurthy, Yuchao Song
  • Patent number: 9865265
    Abstract: A speech recognition system for resolving impaired utterances can have a speech recognition engine configured to receive a plurality of representations of an utterance and concurrently to determine a plurality of highest-likelihood transcription candidates corresponding to each respective representation of the utterance. The recognition system can also have a selector configured to determine a most-likely accurate transcription from among the transcription candidates. As but one example, the plurality of representations of the utterance can be acquired by a microphone array, and beamforming techniques can generate independent streams of the utterance across various look directions using output from the microphone array.
    Type: Grant
    Filed: June 6, 2015
    Date of Patent: January 9, 2018
    Assignee: APPLE INC.
    Inventors: Sean A. Ramprashad, Harvey D. Thornburg, Arvindh Krishnaswamy, Aram M. Lindahl
  • Patent number: 9858944
    Abstract: Apparatus for linear and nonlinear acoustic echo control includes loudspeaker, first, second, and third microphone, beamformer, and first echo canceller. The loudspeaker outputs a loudspeaker signal that includes reference signal. The first microphone and the second microphone are collocated with the loudspeaker, receive at least one of: a near-end speaker signal from a near-end speaker and the loudspeaker signal, and generate first and second microphone uplink signals, respectively. The third microphone receives the near-end speaker signal and generates a third microphone uplink signal. The beamformer receives the first and second microphone uplink signals, directs a beam towards the loudspeaker and drives a null towards the near-end speaker, and generates a beamformer output. The first echo canceler receives the third microphone uplink signal and the beamformer output, and cancels echoes in the third microphone uplink signal based on the beamformer output to generate an echo cancelled signal.
    Type: Grant
    Filed: July 8, 2016
    Date of Patent: January 2, 2018
    Assignee: APPLE INC.
    Inventors: Sarmad Aziz Malik, Arvindh Krishnaswamy
  • Patent number: 9672821
    Abstract: Systems and methods for speech recognition system having a speech processor that is trained to recognize speech by considering (1) a raw microphone signal that includes an echo signal and (2) different types of echo information signals from an echo cancellation system (and optionally different types of ambient noise suppression signals from a noise suppressor). The different types of echo information signals may include those used for echo cancelation and those having echo information. The speech recognition system may convert the raw microphone signal and different types of echo information signals (and optional noise suppression signals) into spectral features in the form of a vector, and a concatenator to combine the feature vectors into a total vector (for a period of time) that is used to train the speech processor, and during use of the speech processor to recognize speech.
    Type: Grant
    Filed: August 25, 2015
    Date of Patent: June 6, 2017
    Assignee: Apple Inc.
    Inventors: Arvindh Krishnaswamy, Charles P. Clark, Sarmad Malik
  • Patent number: 9672843
    Abstract: Method of improving audio signal in the spectral domain starts by receiving audio signal that includes signals from sources including speech source and music source. Audio signal is tuned for output by sound output device. Portions of audio signal are analyzed in a spectral domain to determine whether adjustments are required. Analyzing portions of audio signal includes determining whether anomaly is present in frequency band of audio signal in spectral domain by using at least one metric. Metrics include band energy ratios, spectral centroid, spectral tilt, spectral flux, spectral variance, absolute thresholds, and relative thresholds. Audio signal is adjusted to improve audio signal in spectral domain when audio signal is determined to require adjustments. Adjusting audio signal includes adjusting values of the metric in frequency band that is determined to include anomaly to correspond to clustering of metric values for audio signal in spectral domain. Other embodiments are also described.
    Type: Grant
    Filed: September 30, 2014
    Date of Patent: June 6, 2017
    Assignee: Apple Inc.
    Inventors: Arvindh Krishnaswamy, Joseph M. Williams
  • Patent number: 9525392
    Abstract: Method of dynamically adapting playback volume on electronic device starts with processor receiving first user input and first portion of audio content. First user input signals to device to increase or decrease volume of sound output. Processor determines first loudness metric corresponding to first portion of audio content when first user input is received. First loudness metric is measure of loudness of first portion of audio content being outputted. Processor then stores in memory first loudness metric in association with first user input. Memory stores history of loudness metrics in association with user inputs. Processor then determines second loudness metric that is measure of loudness of second portion of audio content that is received and determines second user input associated with second loudness metric using history. Processor generates control signal to automatically control volume of sound output by device corresponding to second user input. Other embodiments are also described.
    Type: Grant
    Filed: January 21, 2015
    Date of Patent: December 20, 2016
    Assignee: Apple Inc.
    Inventor: Arvindh Krishnaswamy
  • Publication number: 20160358602
    Abstract: Systems and methods for speech recognition system having a speech processor that is trained to recognize speech by considering (1) a raw microphone signal that includes an echo signal and (2) different types of echo information signals from an echo cancellation system (and optionally different types of ambient noise suppression signals from a noise suppressor). The different types of echo information signals may include those used for echo cancelation and those having echo information. The speech recognition system may convert the raw microphone signal and different types of echo information signals (and optional noise suppression signals) into spectral features in the form of a vector, and a concatenator to combine the feature vectors into a total vector (for a period of time) that is used to train the speech processor, and during use of the speech processor to recognize speech.
    Type: Application
    Filed: August 25, 2015
    Publication date: December 8, 2016
    Inventors: Arvindh Krishnaswamy, Charles P. Clark, Sarmad Malik
  • Publication number: 20160358619
    Abstract: A speech recognition system for resolving impaired utterances can have a speech recognition engine configured to receive a plurality of representations of an utterance and concurrently to determine a plurality of highest-likelihood transcription candidates corresponding to each respective representation of the utterance. The recognition system can also have a selector configured to determine a most-likely accurate transcription from among the transcription candidates. As but one example, the plurality of representations of the utterance can be acquired by a microphone array, and beamforming techniques can generate independent streams of the utterance across various look directions using output from the microphone array.
    Type: Application
    Filed: June 6, 2015
    Publication date: December 8, 2016
    Inventors: Sean A. Ramprashad, Harvey D. Thornburg, Arvindh Krishnaswamy, Aram M. Lindahl
  • Publication number: 20160358606
    Abstract: A speech recognition system for resolving impaired utterances can have a speech recognition engine configured to receive a plurality of representations of an utterance and concurrently to determine a plurality of highest-likelihood transcription candidates corresponding to each respective representation of the utterance. The recognition system can also have a selector configured to determine a most-likely accurate transcription from among the transcription candidates. As but one example, the plurality of representations of the utterance can be acquired by a microphone array, and beamforming techniques can generate independent streams of the utterance across various look directions using output from the microphone array.
    Type: Application
    Filed: June 6, 2015
    Publication date: December 8, 2016
    Inventors: Sean A. Ramprashad, Harvey D. Thornburg, Arvindh Krishnaswamy, Aram M. Lindahl
  • Patent number: 9508357
    Abstract: Apparatus for optimizing beamformers for echo control comprises microphones to receive acoustic signals, echo cancellers (ECs) respectively coupled to the microphones to adaptively cancel echo in the acoustic signals and to generate EC-acoustic signals, and a first fixed beamformer coupled to the ECs to receive the EC-acoustic signals. The null of the first beamformer is steered in a direction of a first environmental noise source that is determined offline by exciting the ECs with normal speech signals and audio playback signals to cause the ECs to generate test EC-acoustic signals, and selecting the first environmental noise source based on loudness weighted centroids of noise in the test EC-acoustic signals. Apparatus may also include a residual echo suppressor coupled to the first fixed beamformer to perform echo suppression on output of the first fixed beamformer and to generate clean signal. Other embodiments are also described.
    Type: Grant
    Filed: November 21, 2014
    Date of Patent: November 29, 2016
    Assignee: Apple Inc.
    Inventor: Arvindh Krishnaswamy
  • Publication number: 20160294343
    Abstract: Automatic gain control systems disclosed herein can incorporate a confidence metric that can estimate the accuracy of gain adjustments calculated by an automatic gain control module. The confidence metric may be based on a percentage of valid audio samples in a given period of time. Based on the confidence metric, the AGC response may be reduced, delayed, frozen, or otherwise altered from the baseline gain adjustment. Time-averaging process may be used to estimate the input signal power level and determine an appropriate baseline gain adjustment. Additionally, weighting functions can be adjusted to prevent overestimation of the signal power.
    Type: Application
    Filed: June 7, 2016
    Publication date: October 6, 2016
    Inventors: Arvindh Krishnaswamy, Juha O. Merimaa, Kapil Krishnamurthy, Yuchao Song
  • Patent number: 9462381
    Abstract: A multi-band audio compressor that may provide not only better and brighter sound, but also speaker protection. The multi-band audio compressor breaks an input audio signal into different frequency bands. For each band signal, a volume re-mapper translates a user preference volume level to a converted volume level based on a programmable volume curve for the band signal. For each frequency band, the band signal is processed by a gain stage and a compressor. Each gain stage applies a signal gain to the band signal based on the converted volume level. Each compressor compresses the output of the gain stage. After compression, the different frequency band signals are re-combined and the combined audio signal may then be passed to a power amplifier that is driving a speaker. Other embodiments are also described and claimed.
    Type: Grant
    Filed: September 24, 2014
    Date of Patent: October 4, 2016
    Assignee: Apple Inc.
    Inventors: Arvindh Krishnaswamy, Joseph M. Williams
  • Patent number: 9438195
    Abstract: An equalizer that linearly interpolates between two equalization states when transitioning from one equalization state to the other equalization state is described. The equalizer includes a transfer function generator and an equalization module. Each equalization state is defined or determined based on a set of parameters. The transfer function generator generates a set of interpolated transfer functions by performing linear interpolation on a first equalization state and a second equalization state based on the set of parameters. The linear interpolation is performed on corresponding Z-domain poles and zeros of the transfer functions of the first and second equalization states. The equalization module applies the set of interpolated transfer functions generated by the transfer function generator to an input audio signal.
    Type: Grant
    Filed: September 30, 2014
    Date of Patent: September 6, 2016
    Assignee: Apple Inc.
    Inventors: Arvindh Krishnaswamy, Joseph M. Williams
  • Patent number: 9438194
    Abstract: Method of dynamically adapting user volume input range on mobile device having global volume range starts by receiving a volume input selection from a user that is level included in user volume input range. User volume input range is a portion of global volume range. Device's processor then detects ambient noise level surrounding device and adjusts user volume input range from current portion of global volume range to different portion of global volume range based on detected ambient noise level. Volume input selection remains at the same level included in user volume input range after user volume input range is adjusted. Processor may identify sound profile that corresponds to ambient noise level being detected and adjusts user volume input range to a different portion of the global volume range based on identified sound profile. Other embodiments are also described.
    Type: Grant
    Filed: September 30, 2014
    Date of Patent: September 6, 2016
    Assignee: Apple Inc.
    Inventors: Arvindh Krishnaswamy, Joseph M. Williams
  • Patent number: 9432787
    Abstract: Systems and methods for determining the operating condition of multiple microphones of an electronic device are disclosed. A system can include a plurality of microphones operative to receive signals, a microphone condition detector, and a plurality of microphone condition determination sources. The microphone condition detector can determine a condition for each of the plurality of microphones by using the received signals and accessing at least one microphone condition determination source.
    Type: Grant
    Filed: February 9, 2016
    Date of Patent: August 30, 2016
    Assignee: Apple Inc.
    Inventors: Arvindh Krishnaswamy, David T. Yeh, Juha O. Merimaa, Sean A. Ramprashad
  • Publication number: 20160219386
    Abstract: Systems and methods for determining the operating condition of multiple microphones of an electronic device are disclosed. A system can include a plurality of microphones operative to receive signals, a microphone condition detector, and a plurality of microphone condition determination sources. The microphone condition detector can determine a condition for each of the plurality of microphones by using the received signals and accessing at least one microphone condition determination source.
    Type: Application
    Filed: February 9, 2016
    Publication date: July 28, 2016
    Inventors: Arvindh Krishnaswamy, David T. Yeh, Juha O. Merimaa, Sean A. Ramprashad