66 results about "Hidden Markov model" patented technology

For machine segmenting of speech, first utterances from a database of known spoken words are classified and segmented into three broad phonetic classes (BPC) voiced, unvoiced, and silence. Next, using preliminary segmentation positions as anchor points, sequence-constrained vector quantization is used for further segmentation into phoneme-like units. Finally, exact tuning to the segmented phonemes is done through Hidden-Markov Modelling and after training a diphone set is composed for further usage.

A process and device for detecting colon cancer by classifying and annotating clinical features in video data containing colonoscopic features by applying a probabilistic analysis to intra-frame and inter-frame relationships between colonoscopic features in spatially and temporally neighboring portions of video frames, and classifying and annotating as clinical features any of the colonoscopic features that satisfy the probabilistic analysis as clinical features. Preferably the probabilistic analysis is Hidden Markove Model analysis, and the process is carried out by a computer trained using semi supervised learning from labeled and unlabeled examples of clinical features in video containing colonoscopic features.

The automatic Arabic text image optical character recognition method includes training a text recognition system using Arabic printed text, using the produced models for classification of newly unseen Arabic scanned text, and generating the corresponding textual information. Scanned images of Arabic text and copies of minimal Arabic text are used in the training sessions. Each page is segmented into lines. Features of each line are extracted and input to Hidden Markov Model (HMM). All training data training features are used. HMM runs training algorithms to produce codebook and language models. In the classification stage new Arabic text is input in scanned form. Line segmentation where lines are extracted is passed through. In the feature stage, line features are extracted and input to the classification stage. In the classification stage the corresponding Arabic text is generated.

A recognition system (10) incorporates a filterbank analyser (16) producing successive data vectors of energy values for twenty-six frequency intervals in a speech signal. A unit (18) compensates for spectral distortion in each vector. Compensated vectors undergo a transformation into feature vectors with twelve dimensions and are matched with hidden Markov model states in a computer (24). Each matched model state has a mean value which is an estimate of the speech feature vector. A match inverter (28) produces an estimate of the speech data vector in frequency space by a pseudo-inverse transformation. It includes information which will be lost in a later transformation to frequency space. The estimated data vector is compared with its associated speech signal data vector, and infinite impulse response filters (44) average their difference with others. Averaged difference vectors so produced are used by the unit (18) in compensation of speech signal data vectors.

A method of recognizing a radar target comprises producing a sequence of Doppler spectra of radar returns form a scene and producing therefrom a sequence of Doppler feature vectors for a target in the scene. Hidden Markov modelling (HMM) is then used to identify the sequence of Doppler feature vectors as indicating a member of a particular class of targets. HMM is used to identify the sequence of Doppler feature vectors by assigning to each feature vector an occurrence probability by selecting a probability distribution or state from a set thereof associated with a class of targets, multiplying the occurrence probabilities together to obtain an overall probability, repeating for other probability distributions in the set to determine a combination of probability distributions giving highest overall probability for that class of target, then repeating for at least one other class of targets and selecting the target class as being that which yields the highest overall occurrence probability.

The invention discloses a text-independent speech conversion system based on HMM model state mapping, which is composed of a data alignment module, a spectrum conversion model generation module, a rhythm conversion model generation module, an online conversion module and a parameter voice synthesizer; wherein, the data alignment module receives the voice parameters of the source and target speakers, and aligns to the input data according to phoneme information to generate state-aligned data pairs; the spectrum conversion model generation module receives the aligned data pairs and establishes a voice spectrum parameter conversion module based on source and target speakers according to the data; the rhythm conversion model generation module receives the aligned data pairs and establishes a voice rhythm parameter conversion module based on source and target speakers according to the data; the online conversion module obtains the converted voice spectrum parameter and rhythm parameter according to the conversion modules generated by the spectrum conversion model generation module and the rhythm conversion model generation module, and voice data of the source speaker for conversion; the parameter voice synthesizer module receives the converted spectrum information and rhythm information from the online conversion module and outputs the converted voice result.

The invention provides a content sensing method based on network stream behaviors. The content sensing method comprises the following steps: acquiring network flow in an outer network environment andextracting an observational characteristic as a training sample; training a model by utilizing the training sample; inputting a network stream with an unknown type into the model and identifying the content; carrying out incremental learning by utilizing recognized network stream flow and historical model parameters; updating the model parameters to ensure the continuity of model classes. According to the content sensing method provided by the invention, a dynamic modeling capability of a hidden Markovmoder model and a powerful nonlinear expression capability of a deep neural network are utilized; an experiment result shows the feasibility of the method and compares performance advantages of an existing technical scheme.

The invention provides a network attack detection method and device. The method comprises the following steps: according to a preset keyword set, a feature module carries out keyword matching on a network session sample, so that features of the network session sample are obtained; a tagging module carries out part-of-speech tagging on the features of the network session sample in a hidden state through a pre-trained hidden Markov model, so that an analysis function is achieved; and a processing module inputs the network session sample, which has undergone the part-of-speech tagging, into a pre-trained classification model for classified processing, so that whether a network attack exists in the network session sample or not is determined. The method and device provided by the invention has the advantages that manual protocol analysis is not needed, so that the technical problems in the prior art that feature extraction from sample data needs to rely on artificial experience and the analysis efficiency is low are solved.

A speech detection system extracts a plurality of features from multiple input streams. In the acoustic model space, the tree of Gaussians in the model is pruned to include the active states. The Gaussians are mapped to Hidden Markov Model states for Viterbi phoneme alignment. Another feature space, such as the energy feature space is combined with the acoustic feature space. In the feature space, the features are combined and principal component analysis decorrelates the features to fewer dimensions, thus reducing the number of features. The Gaussians are also mapped to silence, disfluent phoneme, or voiced phoneme classes. The silence class is true silence and the voiced phoneme class is speech. The disfluent class may be speech or non-speech. If a frame is classified as disfluent, then that frame is re-classified as the silence class or the voiced phoneme class based on adjacent frames.

An analysis system for unknown application layer protocols, which could automatically discover unknown applications existing in a network, and then obtain keywords, attribute values, status codes or type codes representing semantic meaning of each field in each type of unknown application as well as message formats, dialogue rules and status transfer relations of application layer protocols by using cluster analysis and optimal partitioning method based on hidden semi-Markov model. Unknown application analysis result could be used for flow management and safety protection of a network. The system has the following advantages: it avoids difficulties arising from manual discovery and analysis of unknown applications, and improves network management efficiency and responding speed against new types of network attacks.

The invention discloses a medical history transferring system based on voice recognition. The medical history transferring system based on voice recognition is composed of a voice collection unit, a cloud voice analysis and recognition unit, a medical history transferring unit and an output unit. A model is established by the cloud voice analysis and recognition unit through training conducted on voice and a corresponding text and with the application of a tool kit of an HMM, so that a reasonable application side module is obtained; voice collected by the voice collection unit is sent into the application side module to be converted, a result obtained after conversion is transferred in a formatting and standardized mode through the medical history transferring unit, and finally an electronic medical history is output by the output unit. The medical history is transferred through a voice recognition system so that the work efficiency of doctors can be improved, paperless operation of medical history type-in can be achieved, and the environment can be protected.

The presented method comprises: (1) with satellite cloud chart sequence, generating the differential cloud chart sequence for wavelet transformation to select wavelet domain as the bottom data source and information extraction source; (2) according to the space correlation, taking adaptive state block division with EHMM to the target; (3) according to the time correlation, dynamic tracking and scaling motion energy with the cross entropy; (4) constructing the energy function to primary find out the typhoon center by the least energy value on the low-resolution layer. This invention has strong versatility and high location precision.

The present invention relates to automatic human face-recognizing digital video system and belongs to the machine vision technology product. The system hardware consists of video input device, image acquiring card, parallel processor and display screen. The software consists of image acquiring module, cutting module, pre-treatment module, recognition module, parallel processing module, data storing and polling module, display module, alarm module and remote trnamission module. The system adopts the recognition method with implicit Markovian model to recognize static and dynamic face. The present invention can display face image intuitively while automatic recognition. It can be used in the plases which need to recognize human faces automatically.

The present invention relates to a speechmaker identity authentication method based on lip movement in the field of biological characteristics and mode identification technology. Said method includes the following steps: inputting lip image sequence, dividing lip region based on improved fuzzy aggregation algorithm, modeling lip region based on point drive and extracting various lip characteristics based on lip region model.

The hydroacoustic noise detecting and distinguishing method based on wavelet domain includes the first mean value-eliminating and amplitude normalizing treatment of ship noise and marine environment noise, and the subsequent multiple scale domain decomposing of the wavelet domain and establishing implicit wavelet domain Markov model. Based on the characteristic difference between marine environment noise model and ship radiation noise model, one detection method based on Nyman-Pearson criterion is proposed. The likelihood value between the ship noise and the marine environment noise HMT model is used as the inspection statistic for binary hypothesis inspection to raise the ship noise detecting capacity. One characteristic vector describing the detection sample is formed for judging ship noise type, resulting in raised identification rate and robustness.

The present invention relates to a system and a method for quickly and accurately identifying and classifying resistance genes of a plant from a protein or DNA sequence. In order to identify and classify resistance genes of a plant using a hidden marcov model, conceived is a profile matrix made using a protein sequence of a domain which is encoded by the resistance genes, and a system for identifying the domain of the resistance genes using the profile matrix and classifying the resistance genes by domain combination. The present invention enables effective identification and classification of the resistance genes of a plant using the profile matrix and program, of which the nucleotide base sequence or protein sequence is detected.

Two Chinese words extraction based on method semantic structure binding characterizes it balancing the semantic binding strength between words to judge if the being elected alphabetic string can forma word, that is to renew semantic status transfer probability matrix in HMM and output character probability matrix at a status transfer plate with Baum-welch algorithm on the basis of HMM expressing, Chinese vocabulary sementic to get the joint closing states of related sementic probability and semantic sequence expressing semantic binding relationship according to the probability matrix expressing status transfer time places to calculate the word judgement value.

The invention discloses a Parkinson's disease resting state tremor assessment method based on a wearable somatosensory net, belongs to the field of wireless sensor networks and data analysis thereof,and particularly relates to a method for obtaining and identifying the arm tremor state of a Parkinson's disease patient on the basis of the wearable somatosensory net. The attitude angle of the upperarm, the attitude angle of the lower arm and the attitude angle of the wrist are measured to calculate the angle change amount of the elbow joint and the angle change amount of the wrist joint, the characteristics of the angle change amount are extracted, the real-time characteristics of an electromyographic signal are extracted, a hidden Markov model is trained according to characteristic data and a UPDRS (Unified Parkinson's Disease Rating Scale), and a current optimal state sequence is output. The method can provide technical support for evaluating the arm tremor degree of the Parkinson'sdisease patient, and a theoretical foundation is provided for crowds who include Parkinson's disease patients, old people, weak people and the like and need to know the occurrence of the early-phase Parkinson's disease in time.

The invention provides a speech recognition method and a speech recognition device, wherein the method comprises the steps that: a context-dependent HMM(hidden markov model) is adopted when a decoding network is trained, a sil (silence) model is added to a suffix in the decoding network, acoustic contexts of the HMM state before and after the sil model are regulated, and the HMM state skip sequence of a to-be-recognized speech is acquired through the decoding network. Furthermore, a skip to the head part of a linguistic model is added at the end of the linguistic model in the decoding network to simulate the influence of a pause between sentences on the context information of the linguistic model. According to the speech recognition method and the speech recognition device, speech recognition effect is improved.

The invention discloses a robot emotion model generation method, an emotion model and an interactive method. The interactive method comprises the steps of presetting multiple different emotion types according to control or interactive demands, and taking the emotion types as observable state values; obtaining external factors capable of influencing robot emotions, and setting the external factorsto be in a hidden state; through simulation experiments or real person experiments, obtaining an initial state probability matrix and a state transfer probability matrix corresponding to preset different character types; through a forward-backward algorithm, building hidden Markov models corresponding to the different character types; through the hiding state and by adopting an expectation maximization algorithm, solving the hidden Markov models to obtain probabilities of all observable states; and selecting the emotion type corresponding to the observable state with the highest probability asthe emotion of a robot, and obtaining a robot emotion model. According to the scheme, relatively independent emotions can be established for the robot, so that the authenticity and experience of robot and user emotion interaction are improved.

The automatic Arabic text image optical character recognition method includes training a text recognition system using Arabic printed text, using the produced models for classification of newly unseen Arabic scanned text, and generating the corresponding textual information. Scanned images of Arabic text and copies of minimal Arabic text are used in the training sessions. Each page is segmented into lines. Features of each line are extracted and input to Hidden Markov Model (HMM). All training data training features are used. HMM runs training algorithms to produce codebook and language models. In the classification stage new Arabic text is input in scanned form. Line segmentation where lines are extracted is passed through. In the feature stage, line features are extracted and input to the classification stage. In the classification stage the corresponding Arabic text is generated.

Pathogenic effector proteins which include one or more RxLR, dEER, Pexel or analogous motifs are blocked from entry into plant or animal cells by binding one or more of the motifs with a blocking compound which prevents binding of phosphoinositides or other polar lipids to the motifs which is a prerequisite for translocation of the pathogenic effector proteins into the plant or animal cell. The blocking compounds can take a variety of forms including synthetic peptides or the hydrophilic head-groups of phosphoinositides, phosphatidic acids, phospholipids, or sphingolipids. Suitable blocking compounds can be identified by assays demonstrating binding to RxLR, dEER, Pexel or analogous motifs. In addition, pathogenic effector proteins can be identified by analyzing whether they contain structural RxLR motifs using hidden markov modeling.

Embodiments of the present invention disclose a method, system, and computer program product for speech summarization. A computer receives audio and video components from a video conference. The computer determines which participant is speaking based on comparing images of the participants with template images of speaking and non-speaking faces. The computer determines the voiceprint of the speaking participant by applying a Hidden Markov Model to a brief recording of the voice waveform of the participant and associates the determined voiceprint with the face of the speaking participant. The computer recognizes and transcribes the content of statements made by the speaker, determines the key points, and displays them over the face of the participant in the video conference.

The invention provides a pronunciation method of a three-dimensional visual Chinese mandarin pronunciation dictionary with pronunciation being rich in emotion expression ability and relates to the technical field of voice visualization, language teaching, vocal organ animation and facial animation. The method produces the vocal organ animation and produces the facial animation with vivid expressions at the same time. The method has the advantages that based on truly-captured motion data, a physiology motion mechanism of vocal organs and a harden markov model, the built vocal organ animation has coordination and consistency related to the facial animation, and coarticulation phenomena in continuous voice animation can be completely described; data driving models are embedded into physiology models by the utilization of advantages of the physiology models and advantages of the data driving models on describing face partial detail features and reality sense aspects, and the facial animation with high reality sense is generated. An objective performance test and a subjective performance test to a system verify the effectiveness of the pronunciation method in the aspect of intelligent auxiliary language teaching.

A method for reconfiguring the marginalized decode data of hidden Markovian model (HMM) used in speed recognition features that the HMM transfer probability array is used to describe the dynamic characteristics of speech characteristic vector in time domain, the complete variance array is used to describe the relative characteristics between the components of the characteristic vector for Meizi band, and a data reconfiguring algorithm (VITDI) is used to reconfigured "lost vector". It can improve the noise robustness of speech recognition system.

This invention relates to context-dependent Chinese phone identifying and modeling method, which applies initial consonant right-dependent and final sound left dependent modeling method including: a, creating a context-dependent basic modeling unit by relating the initial consonant with the adjacent right final sound and relating the final sound with its adjacent left initial consonant, b, utilizing the state clustering method to train the model parameters to get an initial HMM, c, utilizing the sub-space clustering method to compress the HMM to generate a final model.