Abstract: The present invention provides a process for separating a good quality information signal from a noisy acoustic environment. The separation process uses a set of at least two spaced-apart transducers to capture noise and information components. The transducer signals, which have both a noise and information component, are received into a separation process. The separation process generates one channel that is substantially only noise, and another channel that is a combination of noise and information. An identification process is used to identify which channel has the information component. The noise signal is then used to set process characteristics that are applied to the combination signal to efficiently reduce or eliminate the noise component. In this way, the noise is effectively removed from the combination signal to generate a good qualify information signal. The information signal may be, for example, a speech signal, a seismic signal, a sonar signal, or other acoustic signal.

Abstract: A system for microphone noise reduction includes first and second filter portions and a control processor adapted to adapt the first and second filter portions in response to a one of a plurality of stored vectors. Each stored vector is representative of acoustic transfer functions in accordance with a model of a vehicle and a respective position within the vehicle. A method for processing microphone signals includes selecting a vehicle model, selecting positions within the vehicle model, measuring acoustic response vectors at the positions, storing the response vectors, selecting one of the response vectors, and adapting first and second filter portions in accordance with the selected response vector.

Abstract: A new iterative hierarchical linear regression method for generating a set of linear transforms to adapt HMM speech models to a new environment for improved speech recognition is disclosed. The method determines a new set of linear transforms at an iterative step by Estimate-Maximize (EM) estimation, and then combines the new set of linear transforms with the prior set of linear transforms to form a new merged set of linear transforms. An iterative step may include realignment of adaptation speech data to the adapted HMM models to further improve speech recognition performance.

Abstract: A method is described that corrects incorrect text associated with recognition errors in computer-implemented speech recognition. The method includes the step of performing speech recognition on an utterance to produce a recognition result for the utterance. The command includes a word and a phrase. The method includes determining if a word closely corresponds to a portion of the phrase. A speech recognition result is produced if the word closely corresponds to a portion of the phrase.

Abstract: A method includes (i) measuring first distances between (a) vectors belonging to a set of vectors that represent an utterance and (b) vectors belonging to a set of vectors that represent a template, the measuring being done in accordance with a first order of the utterance vectors a first order of the template vectors, and (ii) measuring second distances between (a) individual vectors belonging to the set of vectors that represent the utterance and (b) individual vectors belonging to the set of vectors that represent the template, the measuring being done in accordance with a second order of the utterance vectors and a second order of the template vectors, and (iii) in which the first template vector order and the second template vector order are different and/or the first utterance vector order and the second utterance vector order are different.

Abstract: A voice recognition device is provided to improve a recognition rate for objective recognition terms on display. The device includes a voice pickup unit for picking up user's voices, a storing unit for storing a plurality of objective recognition terms, a display unit for displaying a designated number of objective recognition terms stored in the storing unit and a voice recognition unit. The voice recognition unit has a weighting section for weighting the objective recognition terms on display more heavily than those not on display, and a calculating section for calculating respective degrees of agreement between the objective recognition terms after the objective recognition terms are weighted and the user's voices are picked up by the voice pickup unit. Based on this calculating result of the degrees of agreement, the voice recognition device recognize the user's voices.

Abstract: An audio decoding apparatus decodes high frequency component signals using a band expander that generates multiple high frequency subband signals from low frequency subband signals divided into multiple subbands and transmitted high frequency encoded information. The apparatus is provided with an aliasing detector and an aliasing remover. The aliasing detector detects the degree of occurrence of aliasing components in the multiple high frequency subband signals generated by the band expander. The aliasing remover suppresses aliasing components in the high frequency subband signals by adjusting the gain used to generate the high frequency subband signals. Thus occurrence of aliasing can be suppressed and the resulting degradation in sound quality can be reduced, even when real-valued subband signals are used in order to reduce the number of operations.

Abstract: A wideband, high quality audio signal is decoded with few calculations at a low bitrate. Unwanted spectrum components accompanying sinusoidal signal injection by a synthesis subband filter built with real-value operations are suppressed by inserting a suppression signal to subbands adjacent to the subband to which the sine wave is injected. This makes it possible to inject a desired sinusoid with few calculations.

Abstract: Certain embodiments of the invention provide a method and system for allocating memory during encoding of an uncompressed information stream. A method for allocating memory during encoding of a uncompressed information stream may include allocating at least a portion of a device memory for storing at least one of an operating data corresponding to the uncompressed information stream and an operating code corresponding to at least one of a plurality of compression algorithms. At least one of the operating data and the operating code may be stored in at least a portion of the allocated device memory. The method may provide selecting at least one of the compression algorithms and encoding at least a portion of the operating data using the selected compression algorithm, resulting in the creation of a compressed information stream.

Abstract: A speech recognition apparatus including an audio cancellation module is disclosed. The module includes an audio input for receiving an audio signal from a microphone. The module also includes at least two audio inputs for receiving audio signals from respective independent audio sources. The audio cancellation module produces a speech signal by canceling two of the independent audio source signals from the microphone signal. A speech recognizer is used to recognize at least part of the speech signal.

Abstract: A method and apparatus for speech processing in a distributed speech recognition system having a front-end and a back-end. The speech processing steps in the front-end are as follows: extracting speech features from a speech signal and normalizing the speech features in order to alter the power of the noise component in the modulation spectrum in relation to the power of the signal component, especially with frequencies above 10 Hz. A low-pass filter is then used to filter the normalized modulation spectrum in order to improve the signal-to-noise ratio (SNR) in the speech signal. The combination of feature vector normalization and low-pass filtering is effective in noise removal, especially in a low SNR environment.

Abstract: The present invention relates to systems and methods for audio processing. For example, the present invention provides systems and methods for creating standardized text for streaming text application from a variety of different speech-to-text software formats and systems and methods for interfacing a stenographic computer with an electronic communications network.

Abstract: Unsupervised speech data is provided to a speech recognizer that recognizes the speech data and outputs a recognition result along with a confidence measure for each recognized word. A task-related acoustic model is generated based on the recognition result, the speech data and the confidence measure. Additional task independent model can be used. The speech data can be weighted by the confidence measure in generating the acoustic model so that only data that has been recognized with a high degree of confidence will weigh heavily in generation of the acoustic model. The acoustic model can be formed from a Gaussian mean and variance of the data.

Abstract: A method for a guided natural language interface includes inputting to a thin client a query, communicating to an interface intermediary, communicating to an interface descriptor data source, generating an interface descriptor, communicating the interface descriptor to the interface intermediary, communicating the interface descriptor to a parser farm, requesting an appropriate parser corresponding to the interface descriptor, assigning an appropriate parser, parsing, communicating a translation from the step of parsing, to the interface intermediary, and communicating the translation to the thin client. The thin client can be geographically remote from any or all of the steps other than the step of inputting, such that the method is performed over a disparate enterprise, such as a network, for example, the Internet.

Abstract: An audio encoder regulates quality and bitrate with a control strategy. The strategy includes several features. First, an encoder regulates quantization using quality, minimum bit count, and maximum bit count parameters. Second, an encoder regulates quantization using a noise measure that indicates reliability of a complexity measure. Third, an encoder normalizes a control parameter value according to block size for a variable-size block. Fourth, an encoder uses a bit-count control loop de-linked from a quality control loop. Fifth, an encoder addresses non-monotonicity of quality measurement as a function of quantization level when selecting a quantization level. Sixth, an encoder uses particular interpolation rules to find a quantization level in a quality or bit-count control loop. Seventh, an encoder filters a control parameter value to smooth quality. Eighth, an encoder corrects model bias by adjusting a control parameter value in view of current buffer fullness.

Abstract: An unsupervised adaptation method and apparatus are provided that reduce the storage and time requirements associated with adaptation. Under the invention, utterances are converted into feature vectors, which are decoded to produce a transcript and alignment unit boundaries for the utterance. Individual alignment units and the feature vectors associated with those alignment units are then provided to an alignment function, which aligns the feature vectors with the states of each alignment unit. Because the alignment is performed within alignment unit boundaries, fewer feature vectors are used and the time for alignment is reduced. After alignment, the feature vector dimensions aligned to a state are added to dimension sums that are kept for that state. After all the states in an utterance have had their sums updated, the speech signal and the alignment units are deleted. Once sufficient frames of data have been received to perform adaptive training, the acoustic model is adapted.

Abstract: In a speech recognition platform, a masking unit 17 can be utilized to mask noisy content within an audio sample. By masking such noise in a dynamic but predictable manner, valid content can be preserved while largely overcoming the random and detrimental presence of noise. In one embodiment, speech recognition features are extracted pursuant to a hierarchical process that localizes, at least to some extent, some of the resultant features from other resultant features. As a result, noisy or otherwise unreliable information corresponding to the audio sample will not be leveraged unduly across the entire feature set. In another embodiment, an average energy value for processed samples is calculated with individual energy values that are downwardly weighted when such individual energy values are likely representative of noise.

Abstract: In a method for determining speech quality using an objective measure, in order to enhance prediction reliability of the evaluated quality parameters, distortions of the mean spectral envelope are extensively corrected with a weighting function WT(f) before comparing spectral properties. Additionally, the fixed band limits for integration of spectral power density are suppressed and other band limits are searched for instead in a predetermined optimization area in which the resulting spectral intensity representations of the voice signal to be evaluated and the reference voice signal have maximum similarity. The solutions described can supplement known methods and can be incorporated into their structures.

Abstract: This invention relates to a method of recognizing a speech utterance (s) available in spelled form, comprising a processing stage in which a corresponding letter sequence (r) is estimated by means of a letter speech recognition unit (2) based on Hidden Markov Models, and a second processing stage (3) in which the estimated result (r) produced by the first processing stage utilizing a statistical letter sequence model (4) and a statistical model (5) for the speech recognition unit (2) is post-processed, while the dynamic programming method is used during the post-processing. For providing robust and efficient speech recognition procedures for the use of speech signals for system control, a grid structure on which the dynamic programming is based and whose node points are provided for the assignment to accumulated probability values, is converted into a tree structure and that an A* algorithm is used for finding an optimum tree path.

Abstract: The present invention generates a task-dependent acoustic model from a supervised task-independent corpus and further adapted it with an unsupervised task dependent corpus. The task-independent corpus includes task-independent training data which has an acoustic representation of words and a sequence of transcribed words corresponding to the acoustic representation. A relevance measure is defined for each of the words in the task-independent data. The relevance measure is used to weight the data associated with each of the words in the task-independent training data. The task-dependent acoustic model is then trained based on the weighted data for the words in the task-independent training data.