Abstract: Various implementations disclosed herein include a phonotactic post-processor configured to rescore the N-best phoneme candidates output by a primary ensemble phoneme neural network using a priori phonotactic information. In various implementations, one of the scored set of the N-best phoneme candidates is selected as a preferred estimate for a one-phoneme output decision by the phonotactic post-processor. In some implementations, the one-phoneme output decision is an estimate of the most likely detected and recognized phoneme in a frame based on a function of posterior probabilities generated by an ensemble phoneme neural network, as well as phonotactic information and statistical performance characterizations incorporated by the phonotactic post-processor.

Abstract: Various implementations disclosed herein include a training module configured to determining a set of detection normalization threshold values associated with speaker dependent voiced sound pattern (VSP) detection. In some implementations, a method includes obtaining segment templates characterizing a concurrent segmentation of a first subset of a plurality of vocalization instances of a VSP, each segment template provides a stochastic characterization of how a particular portion of the VSP is vocalized by a particular speaker; generating a noisy segment matrix using a second subset of the plurality of vocalization instances of the VSP, wherein the noisy segment matrix includes one or more noisy copies of segment representations of the second subset; scoring segments from the noisy segment matrix against the segment templates; and determining detection normalization threshold values at two or more known SNR levels for at least one particular noise type based on a function of the scoring.

Abstract: The various implementations described enable voice activity detection and/or pitch estimation for speech signal processing in, for example and without limitation, hearing aids, speech recognition and interpretation software, telephony, and various applications for smartphones and/or wearable devices. In particular, some implementations include systems, methods and/or devices operable to detect voice activity in an audible signal by determining a voice activity indicator value that is a normalized function of signal amplitudes associated with at least two sets of spectral locations associated with a candidate pitch. In some implementations, voice activity is considered detected when the voice activity indicator value breaches a threshold value. Additionally and/or alternatively, in some implementations, analysis of the audible signal provides a pitch estimate of detectable voice activity.

Abstract: Various implementations disclosed herein include a training module configured to produce a set of segment templates from a concurrent segmentation of a plurality of vocalization instances of a VSP vocalized by a particular speaker, who is identifiable by a corresponding set of vocal characteristics. Each segment template provides a stochastic characterization of how each of one or more portions of a VSP is vocalized by the particular speaker in accordance with the corresponding set of vocal characteristics. Additionally, in various implementations, the training module includes systems, methods and/or devices configured to produce a set of VSP segment maps that each provide a quantitative characterization of how respective segments of the plurality of vocalization instances vary in relation to a corresponding one of a set of segment templates.

Abstract: Various implementations disclosed herein include a training module configured to concurrently segment a plurality of vocalization instances of a voiced sound pattern (VSP) as vocalized by a particular speaker, who is identifiable by a corresponding set of vocal characteristics. Aspects of various implementations are used to determine a concurrent segmentation of multiple similar instances of a VSP using a modified hierarchical agglomerative clustering (HAC) process adapted to jointly and simultaneously segment multiple similar instances of the VSP. Information produced from multiple instances of a VSP vocalized by a particular speaker characterize how the particular speaker vocalizes the VSP and how those vocalizations may vary between instances. In turn, in some implementations, the information produced using the modified HAC process is sufficient to determine more a reliable detection (and/or matching) threshold metric(s) for detecting and matching the VSP as vocalized by the particular speaker.

Abstract: Various implementations disclosed herein include an expert-assisted phoneme recognition neural network system configured to recognize phonemes within continuous large vocabulary speech sequences without using language specific models (“left-context”), look-ahead (“right-context”) information, or multi-pass sequence processing, and while operating within the resource constraints of low-power and real-time devices. To these ends, in various implementations, an expert-assisted phoneme recognition neural network system as described herein utilizes a-priori phonetic knowledge. Phonetics is concerned with the configuration of the human vocal tract while speaking and acoustic consequences on vocalizations. While similar sounding phonemes are difficult to detect and are frequently misidentified by previously known neural networks, phonetic knowledge gives insight into what aspects of sound acoustics contain the strongest contrast between similar sounding phonemes.

Abstract: Various implementations disclosed herein include an expert-assisted phoneme recognition neural network system configured to recognize phonemes within continuous large vocabulary speech sequences without using language specific models (“left-context”), look-ahead (“right-context”) information, or multi-pass sequence processing, and while operating within the resource constraints of low-power and real-time devices. To these ends, in various implementations, an expert-assisted phoneme recognition neural network system as described herein utilizes a-priori phonetic knowledge. Phonetics is concerned with the configuration of the human vocal tract while speaking and acoustic consequences on vocalizations. While similar sounding phonemes are difficult to detect and are frequently misidentified by previously known neural networks, phonetic knowledge gives insight into what aspects of sound acoustics contain the strongest contrast between similar sounding phonemes.

Abstract: Implementations include systems, methods and/or devices operable to enhance the intelligibility of a target speech signal by targeted voice model based processing of a noisy audible signal. In some implementations, an amplitude-independent voice proximity function voice model is used to attenuate signal components of a noisy audible signal that are unlikely to be associated with the target speech signal and/or accentuate the target speech signal. In some implementations, the target speech signal is identified as a near-field signal, which is detected by identifying a prominent train of glottal pulses in the noisy audible signal.

Abstract: Implementations include systems, methods and/or devices operable to detect voice activity in an audible signal by detecting glottal pulses. The dominant frequency of a series of glottal pulses is perceived as the intonation pattern or melody of natural speech, which is also referred to as the pitch. However, as noted above, spoken communication typically occurs in the presence of noise and/or other interference. In turn, the undulation of voiced speech is masked in some portions of the frequency spectrum associated with human speech by the noise and/or other interference. In some implementations, detection of voice activity is facilitated by dividing the frequency spectrum associated with human speech into multiple sub-bands in order to identify glottal pulses that dominate the noise and/or other inference in particular sub-bands. Additionally and/or alternatively, in some implementations the analysis is furthered to provide a pitch estimate of the detected voice activity.

Abstract: Implementations of systems, method and devices described herein enable enhancing the intelligibility of a target voice signal included in a noisy audible signal received by a hearing aid device or the like. In particular, in some implementations, systems, methods and devices are operable to generate a machine readable formant based codebook. In some implementations, the method includes determining whether or not a candidate codebook tuple includes a sufficient amount of new information to warrant either adding the candidate codebook tuple to the codebook or using at least a portion of the candidate codebook tuple to update an existing codebook tuple. Additionally and/or alternatively, in some implementations systems, methods and devices are operable to reconstruct a target voice signal by detecting formants in an audible signal, using the detected formants to select codebook tuples, and using the formant information in the selected codebook tuples to reconstruct the target voice signal.

Abstract: Various implementations described herein include directional filtering of audible signals, which is provided to enable acoustic isolation and localization of a target voice source. Without limitation, various implementations are suitable for speech signal processing applications in, hearing aids, speech recognition software, voice-command responsive software and devices, telephony, and various other applications associated with mobile and non-mobile systems and devices. In particular, some implementations include systems, methods and/or devices operable to emphasize at least some of the time-frequency components of an audible signal that originate from a target direction and source, and/or deemphasizing at least some of the time-frequency components that originate from one or more other directions or sources. In some implementations, directional filtering includes applying a gain function to audible signal data received from multiple audio sensors.

Abstract: Various implementations described herein include directional filtering of audible signals, which is provided to enable acoustic isolation and localization of a target voice source. Without limitation, various implementations are suitable for speech signal processing applications in, hearing aids, speech recognition software, voice-command responsive software and devices, telephony, and various other applications associated with mobile and non-mobile systems and devices. In particular, some implementations include systems, methods and/or devices operable to emphasize at least some of the time-frequency components of an audible signal that originate from a target direction and source, and/or deemphasizing at least some of the time-frequency components that originate from one or more other directions or sources. In some implementations, directional filtering includes applying a gain function to audible signal data received from multiple audio sensors.

Abstract: The various implementations described enable voice activity detection and/or pitch estimation for speech signal processing in, for example and without limitation, hearing aids, speech recognition and interpretation software, telephony, and various applications for smartphones and/or wearable devices. In particular, some implementations include systems, methods and/or devices operable to detect voice activity in an audible signal by determining a voice activity indicator value that is a normalized function of signal amplitudes associated with at least two sets of spectral locations associated with a candidate pitch. In some implementations, voice activity is considered detected when the voice activity indicator value breaches a threshold value. Additionally and/or alternatively, in some implementations, analysis of the audible signal provides a pitch estimate of detectable voice activity.

Abstract: Implementations of systems, method and devices described herein enable enhancing the intelligibility of a target voice signal included in a noisy audible signal received by a hearing aid device or the like. In particular, in some implementations, systems, methods and devices are operable to generate a machine readable formant based codebook. In some implementations, the method includes determining whether or not a candidate codebook tuple includes a sufficient amount of new information to warrant either adding the candidate codebook tuple to the codebook or using at least a portion of the candidate codebook tuple to update an existing codebook tuple. Additionally and/or alternatively, in some implementations systems, methods and devices are operable to reconstruct a target voice signal by detecting formants in an audible signal, using the detected formants to select codebook tuples, and using the formant information in the selected codebook tuples to reconstruct the target voice signal.

Abstract: Implementations of systems, method and devices described herein enable enhancing the intelligibility of a target voice signal included in a noisy audible signal received by a hearing aid device or the like. In particular, in some implementations, systems, methods and devices are operable to generate a machine readable formant based codebook. In some implementations, the method includes determining whether or not a candidate codebook tuple includes a sufficient amount of new information to warrant either adding the candidate codebook tuple to the codebook or using at least a portion of the candidate codebook tuple to update an existing codebook tuple. Additionally and/or alternatively, in some implementations systems, methods and devices are operable to reconstruct a target voice signal by detecting formants in an audible signal, using the detected formants to select codebook tuples, and using the formant information in the selected codebook tuples to reconstruct the target voice signal.