Abstract: A Natural Language Understanding system is provided for indexing of free text documents. The system according to the invention utilizes typographical and functional segmentation of text to identify those portions of free text that carry meaning. The system then uses words and multi-word terms and phrases identified in the free to text to identify concepts in the free text. The system uses a lexicon of terms linked to a formal ontology that is independent of a specific language to extract concepts from the free text based on the words and multi-word terms in the free text. The formal ontology contains both language independent domain knowledge concepts and language dependent linguistic concepts that govern the relationships between concepts and contain the rules about how language works. The system according to the current invention may preferably be used to index medical documents and assign codes from independent coding systems, such as, SNOMED, ICD-9 and ICD-10.

Abstract: Methods, systems, and apparatus, including computer program products, in which data from web documents are partitioned into a training corpus and a development corpus are provided. First word probabilities for words are determined for the training corpus, and second word probabilities for the words are determined for the development corpus. Uncertainty values based on the word probabilities for the training corpus and the development corpus are compared, and new words are identified based on the comparison.

Abstract: A method for training a spoken language identification system to identify an unknown language as one of a plurality of known candidate languages includes the process of creating a sound inventory comprising a plurality of sound tokens, the collective plurality of sound tokens provided from a subset of the known candidate languages. The method further includes providing a plurality of training samples, each training sample composed within one of the known candidate languages. Further included is the process of generating one or more training vectors from each training database, wherein each training vector is defined as a function of said plurality of sound tokens provided from said subset of the known candidate languages. The method further includes associating each training vector with the candidate language of the corresponding training sample.

Abstract: Context-based word prediction is provided. A software application utilizes words contained in an application document to provide context-based word prediction in the same or a related document. The software application creates an application defined data source and populates the data source with words occurring in a document. When the same or a related document is being edited via an input method, for example, typing, speech recognition, electronic handwriting, etc., a prediction engine presents candidate words from the application defined data source that match current text input, and the user may choose from the presented candidate words for automatic population into the document being edited. Information from the application defined data source may be transferred between computing devices, for example, between a mobile computing device and a desktop (non-mobile) computing device.

Abstract: An apparatus and method that integrates both phrase-based and free-form speech-to-speech translation approaches using probability models. The starting step of the method is to receive vocal communication in a source language. Then store the received vocal communication. Then decipher the content of the vocal communication. Then locate in a multilingual dictionary module the corresponding translation of the deciphered vocal communication provided a preset sentence exists in a speech recognition module for the vocal communication. Then translate the vocal communication into the target language provided there is no corresponding translation located in the multilingual dictionary module. Then synthesize the translated target language when there is no corresponding translation for the vocal communication in the multilingual dictionary module. Then store the sound of the translated target language. Then play the sound of the translated target language.

Abstract: A scalable encoding device for realizing scalable encoding by CELP encoding of a stereo sound signal and improving the encoding efficiency. In this device, an adder and a multiplier obtain an average of a first channel signal CH1 and a second channel signal CH2 as a monaural signal M. A CELP encoder for a monaural signal subjects the monaural signal M to CELP encoding, outputs the obtained encoded parameter to outside, and outputs a synthesized monaural signal M? synthesized by using the encoded parameter to a first channel signal encoder. By using the synthesized monaural signal M? and the second channel signal CH2, the first channel signal encoder subjects the first channel signal CH1 to CELP encoding to minimize the sum of the encoding distortion of the first channel signal CH1 and the encoding distortion of the second channel signal CH2.

Abstract: According to an embodiment, voice recognition apparatus includes units of: acoustic processing, voice interval detecting, dictionary, collating, search target selecting, storing and determining, and voice recognition method includes processes of: selecting a search range on basis of a beam search, setting and storing a standard frame, storing an output probability of a certain transition path, determining whether or not the output probability of a certain path is stored. Number of times of calculation of the output probability is reduced by selecting the search range on basis of the beam search, calculating the output probability of the certain transition path only once in an interval from when the standard frame is set to when the standard frame is renewed, and storing and using thus calculated value as an approximate value of the output probability in subsequent frames.

Abstract: A method, a system, and an apparatus for efficiently presenting correction options. The present invention is capable of analyzing user voice commands and sorting multiple input requests based on user selection probability to determine whether a confirmation step should be presented and, if so, the manner in which the confirmation step should be presented. In particular, the method requests an information input from the user and then assigns a confidence level to the information input. If the confidence level is LOW, then the system performs an immediate confirmation step. If the confidence level assigned is MEDIUM or HIGH, then the information is placed into a data set that is confirmed in a batch confirmation step. The batch confirmation step presents the captured information to the user for confirmation. If any of the information is incorrect, then the method sorts the information in ascending order by confidence level and creates a menu of items that may be changed. The user then makes the change.

Abstract: The present invention provides a method and apparatus for enrollment and verification of speaker authentication. The method for enrollment of speaker authentication, comprising: extracting an acoustic feature vector sequence from an enrollment utterance of a speaker; and generating a speaker template using the acoustic feature vector sequence; wherein said step of extracting an acoustic feature vector sequence comprises: generating a filter-bank for the enrollment utterance of the speaker for filtering locations and energies of formants in the spectrum of the enrollment utterance based on the enrollment utterance; filtering the spectrum of the enrollment utterance by the generated filter-bank; and generating the acoustic feature vector sequence from the filtered enrollment utterance.

Abstract: Systems, methods, and apparatuses, including computer program products, are provided for representing language models. In some implementations, a computer-implemented method is provided. The method includes generating a compact language model including receiving a collection of n-grams from the corpus, each n-gram of the collection having a corresponding first probability of occurring in the corpus and generating a trie representing the collection of n-grams. The method also includes using the language model to identify a second probability of a particular string of words occurring.

Abstract: An audio encoder, an audio decoder or an audio processor includes a filter for generating a filtered audio signal, the filter having a variable warping characteristic, the characteristic being controllable in response to a time-varying control signal, the control signal indicating a small or no warping characteristic or a comparatively high warping characteristic. Furthermore, a controller is connected for providing the time-varying control signal, which depends on the audio signal. The filtered audio signal can be introduced to an encoding processor having different encoding algorithms, one of which is a coding algorithm adapted to a specific signal pattern. Alternatively, the filter is a post-filter receiving a decoded audio signal.

Abstract: A motor vehicle has a speech interface for an acoustic input of commands for operating the motor vehicle or a module of the motor vehicle. The speech interface includes a speech recognition database in which a substantial portion of commands or command components, which can be input, are stored in a version according to a pronunciation in a first language and in a version according to a pronunciation in at least a second language, and a speech recognition engine for automatically comparing an acoustic command to commands and/or command components, which are stored in the speech recognition database, in a version according to the pronunciation in the first language and to commands and/or command components, which are stored in the speech recognition database, in a version according to the pronunciation in the second language.

Abstract: A transient noise removal system removes or dampens undesired transients from speech. When the transient noise removal system receives a speech frame, the system performs a wavelet transform analysis. The speech frame may be represented by one or more wavelet coefficients across one or more wavelet levels. For a given wavelet level, the transient noise-removal system may determine a wavelet threshold. The transient noise removal system may compare the threshold corresponding to a wavelet level to the wavelet coefficients within that level. The transient noise removal system may attenuate each wavelet coefficient based on a comparison to a threshold.

Abstract: A method and system for teaching a foreign language to a user who has knowledge of a base language is disclosed. The method and system may include delivering a video presentation simultaneously to a plurality of users. The method and system may also include simultaneously delivering a plurality of mixed known language-foreign language audio and/or text streams to the plurality of users, each of the plurality of mixed known language-foreign language audio and/or text streams corresponding to the video presentation.

Abstract: A method for improving speech recognition accuracy includes utilizing skiplists or lists of values that cannot occur because of improbability or impossibility. A table or list is stored in a dialog manager module. The table includes a plurality of information items and a corresponding list of improbable values for each of the plurality of information items. A plurality of recognized ordered interpretations is received from an automatic speech recognition (ASR) engine. Each of the plurality of recognized ordered interpretations each includes a number of information items. A value of one or more of the received information items for a first recognized ordered interpretation is compared to a table to determine if the value of the one of the received information items matches any of the list of improbable values for the corresponding information item.

Abstract: A computer-implemented method for creating a disfluency translation lattice includes providing a plurality of weighted finite state transducers including a translation model, a language model, and a phrase segmentation model as input, performing a cascaded composition of the weighted finite state transducers to create a disfluency translation lattice, and storing the disfluency translation lattice to a computer-readable media.

Abstract: In a QA (Question/Answer) system, candidate answers in response to a question received are ranked by probabilities estimated by a language model. The language model is created based on an ordered centroid created from the question and information learned from an information source such as the Internet.

Abstract: A scalable encoding apparatus, a scalable decoding apparatus and the like are disclosed which can achieve a band scalable LSP encoding that exhibits both a high quantization efficiency and a high performance. In these apparatuses, a narrow band-to-wide band converter receives and converts a quantized narrow band LSP to a wide band, and then outputs the quantized narrow band LSP as converted (i.e., a converted wide band LSP parameter) to an LSP-to-LPC converter. The LSP-to-LPC converter converts the quantized narrow band LSP as converted to a linear prediction coefficient and then outputs it to a pre-emphasizer. The pre-emphasizer calculates and outputs the pre-emphasized linear prediction coefficient to an LPC-to-LSP converter. The LPC-to-LSP converter converts the pre-emphasized linear prediction coefficient to a pre-emphasized quantized narrow band LSP as wide band converted, and then outputs it to a prediction quantizer.

Abstract: An apparatus and method of low bit rate encoding and reproducing. The method includes transforming input audio signals in a time domain into spectral signals in a frequency domain, extracting important-spectrum components from the spectral signals in the frequency domain, and quantizing the important-spectrum components, extracting residual-spectrum components other than the important-spectrum components from the spectral signals in the frequency domain, and calculating and quantizing a noise level of the residual-spectrum components, and encoding the quantized important-spectrum components and the quantized noise level losslessly, and outputting encoded bitstreams.

Abstract: The present invention relates to encoding technology. The encoding method includes selecting a second encoding mode for encoding an input frame signal according to an analysis on signal characteristic of the input frame signal; obtaining coding demand values for a preset first encoding mode and the second encoding mode which are used to encode the input frame signal; determining, from the above encoding modes based on the coding demand values, an encoding mode for encoding the input frame signal; and multiplexing information of the determined encoding mode and encoded data which are encoded according to the determined encoding mode. Hence, the compatibility and the prioritization in terms of the encoding modes can be achieved.