Abstract: This invention provides a device and method for language model switching and adaptation, wherein the device comprises a notification manager which notifies a language model switching section of the current status information or the request for the language model of an destination application when the status of the destination application is changed; a language model switching section which selects one or more language models to be switched from a language model set according to the received current status information or the request; a LMB engine decodes a user input using the one or more selected language models; and a language model adaptation section which receives the decoded result and modifies the one or more selected language models based on the decoded result. Therefore, the user input is more accurate even if the language model switching section performs different switches among different language models and the performance of the language models are improved by the language model adaptation section.

Abstract: A system and method for prospecting digital information is provided. A home evergreen index for a home subject area within a corpus of digital information is maintained and includes topic models matched to the corpus. A frontier evergreen index for a frontier subject area within the corpus topically distinct from the home subject area is identified. Quality assessments for frontier articles from the corpus identified by the topic models of the frontier evergreen index are obtained. The frontier articles with positive quality assessments are reclassified against the topic models in the home evergreen index. The frontier articles are provided in a display with home articles previously classified against the topic models in the home evergreen index.

Abstract: An audio encoder and decoder use architectures and techniques that improve the efficiency of multi-channel audio coding and decoding. The described strategies include various techniques and tools, which can be used in combination or independently. For example, an audio encoder performs a pre-processing multi-channel transform on multi-channel audio data, varying the transform so as to control quality. The encoder groups multiple windows from different channels into one or more tiles and outputs tile configuration information, which allows the encoder to isolate transients that appear in a particular channel with small windows, but use large windows in other channels. Using a variety of techniques, the encoder performs flexible multi-channel transforms that effectively take advantage of inter-channel correlation. An audio decoder performs corresponding processing and decoding. In addition, the decoder performs a post-processing multi-channel transform for any of multiple different purposes.

Abstract: Content matching using phoneme comparison and scoring is described, including extracting phonemes from a file, comparing the phonemes to other phonemes, associating a first score with the phonemes based on a probability of the other phonemes matching the phonemes, and providing the file with another file when a request is received to access one or more files having a second score that is substantially similar to the first score.

Abstract: A word alignment apparatus includes a word extracting portion that extracts each word from an example sentence and from a translation sentence thereof, an alignment calculator that calculates at least one of a similarity degree and an association degree between a word in a first language and that in a second language to perform an alignment between words respectively included in the example sentence in the first language and those included in the translation sentence thereof in the second language on the basis of a calculated value, and an optimization portion that optimizes the alignment by performing a bipartite graph matching.

Abstract: An audio encoder and decoder use architectures and techniques that improve the efficiency of quantization (e.g., weighting) and inverse quantization (e.g., inverse weighting) in audio coding and decoding. The described strategies include various techniques and tools, which can be used in combination or independently. For example, an audio encoder quantizes audio data in multiple channels, applying multiple channel-specific quantizer step modifiers, which give the encoder more control over balancing reconstruction quality between channels. The encoder also applies multiple quantization matrices and varies the resolution of the quantization matrices, which allows the encoder to use more resolution if overall quality is good and use less resolution if overall quality is poor. Finally, the encoder compresses one or more quantization matrices using temporal prediction to reduce the bitrate associated with the quantization matrices. An audio decoder performs corresponding inverse processing and decoding.

Abstract: A system and apparatus for establishing whether a received signal frame is an audio signal frame is disclosed. In one embodiment, the system includes a predetermined position in an audio signal frame containing a piece of secondary information for an audio characteristic of the audio data, with a selection device for selecting a succession of bits which is arranged at the predetermined position in the received signal frame. A decision-making device flags the received signal frame as an audio signal frame if the succession of bits represents the piece of secondary information.

Abstract: In one embodiment, the method includes receiving the audio signal including a block of audio data partitioned into N sub-blocks, and restoring a plurality of code parameters s(0), s(1), . . . , s(N?1), respectively. The restoring step includes detecting s(0) from the audio signal, where s(0) represents the code parameter of the first sub-block; detecting a difference s(i)?s(i?1) from the audio signal for i=1, . . . N?1, where s(i) representing the code parameter of each sub-block following the first sub-block. The difference s(i)?s(i?1) is encoded by using first entropy code. The restoring step further includes calculating s(i) for i=1, . . . , N?1 using s(0) and the detected differences, and the method further includes decoding the N sub-blocks using the restored code parameters.

Abstract: A system and method for identifying contact information is provided. A system to identify contact information may include an input to receive a data stream. The data stream may include audio content, video content or both. The system may also include an analysis module to detect contact information within the data stream. The system may also include a memory to store a record of the contact information.

Abstract: A method and system for parsing an address is disclosed. The method and system comprise separating the address into a plurality of tokens and providing one or more token meaning discovery passes based upon region specific configuration information to determine the meaning of each token in the address. In so doing, an address can be parsed by a processing system in an efficient and effective fashion. By disclosing the meaning of each token of the address in accordance with a region specific configuration information rule set a parsing process is provided which allows for easy modification as the requirements for the parsing change.

Abstract: Provided is an apparatus and method for output voice, which receives an information item suitable to a user's taste among information items existing on a network such as the Internet in a text format, converts the information item into voice, and then outputs the voice. The apparatus to output voice includes an information search unit searching at least one first information item corresponding to a preset information class among information items existing on a network, an information processing unit extracting a core information item from the first information items in such a manner as to correspond with a preset reproducing time period, a voice generating unit converting the core information into voice, and an output unit outputting the converted voice.

Abstract: A method of statistical machine translation (SMT) is provided. The method comprises generating reordering knowledge based on the syntax of a source language (SL) and a number of alignment matrices that map sample SL sentences with sample target language (TL) sentences. The method further comprises receiving a SL word string and parsing the SL word string into a parse tree that represents the syntactic properties of the SL word string. The nodes on the parse tree are reordered based on the generated reordering knowledge in order to provide reordered word strings. The method further comprises translating a number of reordered word strings to create a number of TL word strings, and identifying a statistically preferred TL word string as a preferred translation of the SL word string.

Abstract: A method and device for updating statuses of synthesis filters are provided. The method includes: exciting a synthesis filter corresponding to a first encoding rate by using an excitation signal of the first encoding rate, outputting reconstructed signal information, and updating status information of the synthesis filter and a synthesis filter corresponding to a second encoding rate. In the present disclosure, the status of the synthesis filter corresponding to the current rate and the statuses of the synthesis filters at other rates are updated. Thus, synchronization between the statuses of the synthesis filters corresponding to different rates at the encoding terminal may be realized, thereby facilitating the consistency of the reconstructed signals of the encoding and decoding terminals when the encoding rate is switched, and improving the quality of the reconstructed signal of the decoding terminal.

Abstract: An automated name searching system incorporates an automatic name classifier and a multi-path architecture in which different algorithms are applied based on cultural identity of the query name. The name classifier operates with a preemptive list, analysis of morphological elements, length, and linguistic rules. A name regularizer produces a character based computational representation of the name. A pronunciation equivalent representation such as an IPA language representation, and language specific rules to generate name searching keys, are used in a first pass to eliminate database entries which are obviously not matches for the query name. The methods can also be implemented as a callable set of library routines including an intelligent preprocessor and a name evaluator that produces a score comparing a query name and database name, based on a variety of user-adjustable parameters. The user-controlled parameters permit tuning of the search methodologies for specific custom applications.

Abstract: Embodiments of the present invention improve methods of performing speech recognition using sensory inputs of human position. In one embodiment, the present invention includes a speech recognition method comprising sensing a change in position of at least one part of a human body, selecting a recognition set based on the change of position, receiving a speech input signal, and recognizing the speech input signal in the context of the first recognition set.

Abstract: A method and corresponding apparatus for coded-domain acoustic echo control is presented. An echo control problem is considered as that of perceptually matching an echo signal to a reference signal. A perceptual similarity function that is based on the coded spectral parameters produced by the speech codec is defined. Since codecs introduce a significant degree of non-linearity into the echo signal, the similarity function is designed to be robust against such effects. The similarity function is incorporated into a coded-domain echo control system that also includes spectrally-matched noise injection for replacing echo frames with comfort noise. Using actual echoes recorded over a commercial mobile network, it is shown herein that the similarity function is robust against both codec non-linearities and additive noise. Experimental results further show that the echo-control is effective at suppressing echoes compared to a Normalized Least Mean Squared (NLMS)-based echo cancellation system.

Abstract: A generic predictive argument model that can be applied to a set of shot values to predict a target slot value is provided. The generic predictive argument model can predict whether or not a particular value or item is the intended target of the user command given various features. A prediction for each of the slot values can then be normalized to infer a distribution over all values or items. For any set of slot values (e.g., contacts), a number of binary variable s are created that indicate whether or not each specific slot value was the intended target. For each slot value, a set of input features can be employed to predict the corresponding binary variable. These input features are generic properties of the contact that are “instantiated” based o n properties of the contact (e.g., contact-specific features). These contact-specific features can be stored in a user data store.

Abstract: A method for speech enabling an application can include the step of specifying a speech input within a speech-enabled markup. The speech-enabled markup can also specify an application operation that is to be executed responsive to the detection of the speech input. After the speech input has been defined within the speech-enabled markup, the application can be instantiated. The specified speech input can then be detected and the application operation can be responsively executed in accordance with the specified speech-enabled markup.

Abstract: Embodiments of the present invention improve speech recognition using historical information. In one embodiment, the present invention includes a method of performing speech recognition comprising receiving an identifier specifying a user of a kiosk, retrieving history information about the user using the identifier, receiving speech input, recognizing said speech input in the context of a first recognition set, resulting in first recognition results, and modifying the first recognition results using the history information.

Abstract: Aspects of a method and system for decoding GSM speech data using redundancy are provided. A decoding algorithm in a frame process may be utilized to generate a bit-sequence for GSM speech data received via a burst process. The decoding algorithm may be a modified Viterbi algorithm, for example. The frame process may comprise verifying a CRC for the bit-sequence and/or decrypting the bit-sequence. In some instances, estimates of the bit-sequence may be fed back to the decoding algorithm. A speech stream that satisfies speech constraints may be generated based on the generated bit-sequence. The speech constraints may comprise gain and/or pitch continuity, for example. The generated speech stream may be decoded via a voice decoder that supports full rate (FR), adaptive multi-rate (AMR), and/or enhanced full-rate (EFR) speech coding. Frame process results may be fed back to the burst process to improve decoding of received GSM speech data.