Abstract: According to a disclosed embodiment, an endpointer determines the background energy of a first portion of a speech signal, and a cepstral computing module extracts one or more features of the first portion. The endpointer calculates an average distance of the first portion based on the features. Subsequently, an energy computing module measures the energy of a second portion of the speech signal, and the cepstral computing module extracts one or more features of the second portion. Based on the features of the second portion, the endpointer calculates a distance of the second portion. Thereafter, the endpointer contrasts the energy of the second portion with the background energy of the first portion, and compares the distance of the second portion with the distance of the first portion. The second portion of the speech signal is classified by the endpointer as speech or non-speech based on the contrast and the comparison.

Abstract: A process, apparatus, and system are disclosed for inhibiting erroneous control signals due to disruptive electromagnetic signals. In one embodiment, the invention prevents erroneous information writing in a recording device under the effect of background electromagnetic waves. A conductor pattern is formed on a printed wiring board on which wiring constituting a control circuit of a hard disk drive is formed. Background electromagnetic waves are detected by an electromagnetic wave detection circuit. In the case where an electromagnetic wave signal detected by the electromagnetic wave detection circuit exceeds a predetermined power, voltage, or current threshold, a write gate signal is turned off to inhibit a potential write operation.

Abstract: An apparatus and a concomitant method for speech recognition. In one embodiment, the present method is referred to as a “Dynamic Noise Compensation” (DNC) method where the method estimates the models for noisy speech using models for clean speech and a noise model. Specifically, the model for the noisy speech is estimated by interpolation between the clean speech model and the noise model. This approach reduces computational cycles and does not require large memory capacity.

Abstract: A method of content grouping based on a corresponding text includes reading from a content list file for a plurality of content groups, and extracting a key phrase set for each content group. The method includes determining, for each content group, a unique score and a non-unique score, determining a unique threshold and a non-unique threshold, and selecting a content group according to a comparison between the scores and the thresholds.

Abstract: A method and device for controlling a rotation speed of a spindle of an optical disk drive includes a frequency detector, two phase detectors, a frequency divider, a low pass filter, a switch circuit, and a drive circuit. The method uses a wobble signal, an encoder EFM frame synchronization (EEFS) signal, an encoder subcode frame synchronization (ESFS) signal, and other related signals to generate five control signals for controlling the rotation speed of a motor connected to the spindle of the optical disk drive. Using the five control signals, the method limits the phase difference between the Absolute Time in Pre-groove (ATIP) sync and the ESFS to be within a predetermined value, and thus improves an update rate and a writing efficiency.

Abstract: The present invention is a segmenter used in a natural language processing system. The segmenter segments a textual input string into tokens for further natural language processing. In accordance with one feature of the invention, the segmenter includes a tokenizer engine that proposes segmentations and submits them to a linguistic knowledge component for validation. In accordance with another feature of the invention, the segmentation system includes language-specific data that contains a precedence hierarchy for punctuation. If proposed tokens in the input string contain punctuation, they can illustratively be broken into subtokens based on the precedence hierarchy.

Abstract: A system and method are provided for processing audio and speech signals using a pitch and voicing dependent spectral estimation algorithm (voicing algorithm) to accurately represent voiced speech, unvoiced speech, and mixed speech in the presence of background noise, and background noise with a single model. The present invention also modifies the synthesis model based on an estimate of the current input signal to improve the perceptual quality of the speech and background noise under a variety of input conditions. The present invention also improves the voicing dependent spectral estimation algorithm robustness by introducing the use of a Multi-Layer Neural Network in the estimation process. The voicing dependent spectral estimation algorithm provides an accurate and robust estimate of the voicing probability under a variety of background noise conditions. This is essential to providing high quality intelligible speech in the presence of background noise.

Abstract: A data processing system blind source separation of an overcomplete set of signals generally includes means for storing input from sensors in a mixed signal matrix X 200, noise in a noise matrix V 202, an estimate of the individual signals from the mixture of signals from the signal sources in a source signal estimate matrix ? 204, and an estimate of environmental effects in a mixing matrix  206, the matrices related by X=Â?+V; generating an initial estimate of  208; determining the number of, and associated lines of correlation of, each source from Â, and representing the sources in the source signal estimate matrix ? 210; jointly optimizing ? and  in an iterative manner to generate an optimized source signal estimate matrix ? 212 and a final estimated mixing matrix Â; and restoring the separated source signals from the optimized source signal estimate matrix ? 214.

Abstract: A recording clock generating apparatus for a data recording system includes a wobble signal sampler which samples a wobble signal, from wobble grooves of an optical disk. A first phase comparator provides a first phase-difference signal based on a difference in phase between the wobble signal and a first recording clock signal. A prepit detector detects a prepit signal from prepits of the optical disk. A second phase comparator provides a second phase-difference signal based on a difference in phase between the prepit signal and the first recording clock signal. A phase control signal generator provides a phase control signal based on a sum of the first phase-difference signal and the second phase-difference signal. A clock generator provides a second recording clock signal having a phase corrected in response to the phase control signal provided by the phase control signal generator.

Abstract: An arrangement is provided for an automatic speech recognition mechanism to adapt to an adverse acoustic environment. Some of the original training data, collected from an original acoustic environment, is played back in an adverse acoustic environment. The playback data is recorded in the adverse acoustic environment to generate recorded playback data. An existing speech model is then adapted with respect to the adverse acoustic environment based on the recorded playback data and/or the original training data.

Abstract: A speech recognition method and apparatus is disclosed in which each utterance matched to a feature model within a feature model memory (110) is associated with a confidence score indicative of the posterior probability of the word being correctly matched, given that the matching of the utterance to the feature model generated certain values indicative of the goodness of the match. The confidence score for the matching of an utterance to a feature model is determined from the generated values indicative of the goodness of the match and a stored set of parameters indicating the probability of the generated values arising given that a match is either correct or incorrect.

Abstract: In accordance with one aspect of the invention, a selector supports the selection of a first encoding scheme or the second encoding scheme based upon the detection or absence of the triggering characteristic in the interval of the input speech signal. The first encoding scheme has a pitch pre-processing procedure for processing the input speech signal to form a revised speech signal biased toward an ideal voiced and stationary characteristic. The pre-processing procedure allows the encoder to fully capture the benefits of a bandwidth-efficient, long-term predictive procedure for a greater amount of speech components of an input speech signal than would otherwise be possible. In accordance with another aspect of the invention, the second encoding scheme entails a long-term prediction mode for encoding the pitch on a sub-frame by sub-frame basis.

Abstract: A speech recognition system plays prompts to a user in order to obtain information from the user. If the user begins to speak, the prompt should stop. However, the system may receive sounds other than speech from the user while playing a prompt, in which case the prompt should continue. The system temporarily stops a prompt when it detects a sound or when it preliminarily determines that a detected sound may be a target sound (such as words from the user). The system then determines whether the received sound is a target sound or some other sound (such as coughing or a door shutting). If the received sound is not determined to be a target sound, then the prompt is resumed. The prompt can be resumed at any appropriate point, such as the point where it was stopped, a prior phrase boundary, or the beginning of the prompt.

Abstract: A waveform equalizer, which has a partial response characteristic represented in the form of PR (a, b, c, b, a) which is a quaternary transfer function characteristic, is made up an analog filter, an ADC (analog/digital converter), and an FIR filter, for providing matching with the frequency characteristic of a read back waveform read from a recording medium. Such signal processor characteristic approximation to the regenerative signal characteristic makes it possible to easily achieve equalization without particularly emphasizing the regenerative signal, thereby achieving a reduced circuit scale.

Abstract: The type of audio stored in the payload of a data packet transmitted over a data network is identified as speech audio or non-speech audio through the use of a non-speech identifier included in a header in the data packet. Upon detection of data packet containing non-speech audio, the receiver of the data packet may modify jitter buffer latency while the non-speech audio is being received. Modifying the jitter buffer latency while non-speech audio is being received minimizes the loss of spoken words during jitter buffer latency modification.

Abstract: An arrangement is provided for integrating graph decoder with keyword spotting to improve the robustness of speech recognition. When a graph decoder based speech recognition mechanism fails to recognize a word sequence from input speech data, a keyword based speech recognition mechanism is activated to recognize the word sequence based on a set of keywords that are detected from the input data.

Abstract: The present invention is a system and method for computer analysis of computer generated communications to produce indications and warnings of dangerous behavior.

Abstract: There is provided rate selection methods and systems for selecting coding rates for coding frames of a speech signal to realize an average bit rate indicated by a mode. For example, a mode 0, mode 1, and a mode 2 may be defined, with each mode requiring a different average bit rate. To achieve the average bit rate of a particular mode, a coding rate is selected for each frame of the speech signal, based on the characteristics of a frame. A frame can be categorized in a class, such as noise or silence, noise-like unvoiced speech, pulse-like unvoiced speech, transition into voiced speech, unstable voiced speech, stable voiced speech. Other parameters may also be used, such as the sharpness, noise-to-signal ratio, pitch correlation, energy, and reflection coefficient. A frame may then be coded at a full-rate, a half-rate, a quarter-rate, or an eighth-rate.

Abstract: The distance between the first two pitch marks of a voiced portion of speech data to be processed is calculated. The difference between the adjacent inter-pitch-mark distances is calculated. The respective calculation results are stored and managed in a file.

Abstract: A method extracts speech recognition features from a speech signal coded as a bitstream. First, the bitstream is decoded to recover linear predictive coding filter parameters, and a residual signal. Then, the linear predictive coding filter parameters and the residual signal are discriminatively combined into speech recognition features.