7 results about "Sequence memory" patented technology

The invention relates to the technical field of speech recognition, and discloses a speech recognition method and device, equipment and a storage medium. Aiming at the defects of fixed filter coefficient, static context modeling and difficulty in deep training and deployment of the existing deep feed-forward sequence memory network (DFSMN), the method comprises the following steps of: acquiring a frame-level acoustic feature sequence of input voice, and inputting an acoustic model containing a neural network unit; in the unit, the filter weight of the memory module is generated in real time through a parameter generation network based on the intermediate feature of the current frame; performing weighted aggregation on the historical and future frame features by using the weight to obtain a context enhancement feature; and outputting features based on the feature generation unit, and decoding to obtain an identification result. According to the method, context dynamic self-adaptive modeling is realized, the recognition precision and robustness are improved, the efficient reasoning characteristic of a pure feed-forward network is kept, the calculation overhead is not remarkably increased, and the method is adaptive to a server and terminal equipment and is suitable for multi-scene speech recognition of keywords, command words and the like.

The invention discloses a voice control optimization method and system based on big data. The method comprises the steps of voice data acquisition, voice recognition model construction, voice recognition model training and voice intelligent control. The invention relates to the technical field of voice data processing, in particular to a voice control optimization method and system based on big data, which innovatively adopts a three-channel parallel convolution structure, can simultaneously capture local frame association, medium-range dependency and long-range semantic information of voice signals, and improves the discrimination and robustness of voice feature representation; a deep feed-forward sequence memory neural network with improved hierarchical memory size design and interlayer jump connection structure is innovatively introduced, and the voice recognition continuity, context consistency and overall recognition precision are remarkably improved; a semantic understanding model is introduced in a speech recognition model training stage, and a joint loss function is constructed, so that the semantic consistency and speech recognition accuracy of the model in a complex context are remarkably improved.

The invention discloses a low-complexity large-model long-sequence memory modeling method. The method comprises the following steps: firstly, constructing a fractional order state space model FOSSM by introducing a Capto fractional order derivative to replace an integer order differential operator; then, based on an asymptotic behavior of a Mittag-Leffler function, establishing a theoretical relationship between a power law attenuation property of an FOSSM memory kernel and a fractional order parameter; deducing the theoretical advantage boundary of the FOSSM in the aspect of power law data approximation compared with the standard SSM; and finally, the non-rational fractional order operator is converted into a rational transfer function through an Oustaloup rational approximation method, and efficient discretization of calculation complexity is realized. According to the method, power law long-range dependence can be modeled naturally, more matched mathematical representation is provided for a sequence with heavy-tailed time correlation, and the problem of path repetition caused by insufficient historical track memory in unmanned equipment navigation and the problem of poor interaction continuity caused by weak context keeping ability in man-machine interaction in a traditional method are solved.

This invention discloses a low-instruction-overhead, reconfigurable normalization operator implementation method based on SIMD vector processing, relating to deep learning algorithm model technology and electronic information technology. The invention first divides the normalization operator into reduction and scalar operations, compiles the operations into a fixed sequence of M microcodes, preloads them into a microcode sequence memory, and configures a SIMD vector processing unit consisting of an instruction decoding unit, a hardware autonomous execution engine, and a memory access address generator. It continuously performs sequence length token and channel group scanning, performing P×N iterations on P groups of channels and N tokens, physically reusing the microcode sequence. The microcode pointer is automatically reset to zero, and intermediate result fusion is adaptively completed by scalar microcode operator encoding. Compared with existing GPU solutions, this method has better reconfigurability, lower execution cycle, and lower instruction overhead.

The invention discloses a non-intrusive dynamic acquisition method for WeChat 4.0 and above version database keys. The method comprises the following steps: acquiring a key memory address corresponding to a key dynamic library; obtaining a sequence memory address of a key dynamic library where the target instruction machine code sequence is located; obtaining an address relative offset according to the sequence memory address and the first base address; and obtaining a target breakpoint absolute address according to the address relative offset and the first base address, setting a breakpoint according to the target breakpoint absolute address, obtaining address data at the corresponding breakpoint, and performing hexadecimal conversion on the obtained address data to obtain a WeChat key, thereby realizing non-intrusive dynamic obtaining of the database key. According to the method, the problem that the Wechat 4.0 and above version database key cannot be effectively obtained by a traditional key obtaining method is solved.

The application relates to the technical field of wind power intelligent operation and maintenance, and discloses a wind turbine temperature early warning method based on hybrid deep learning and dynamic threshold values, wherein through a hybrid deep learning anomaly detection model, time sequence feature extraction and sequence memory enhancement are performed on synchronous time sequence data, wind turbine fusion features are obtained by fusing fine-grained short-term features and coarse-grained long-term features, the abnormal probability of each moment is predicted, the real-time working condition category is identified, the working condition specificity threshold value is dynamically output, the working condition stability index is combined, and a wind turbine temperature early warning signal is generated. Therefore, based on the hybrid deep learning and dynamic threshold value technology, the short-term peak detail features and the long-term dependence features are fused, the model peak fitting precision is improved, the abnormal probability can accurately reflect the temperature abnormality degree, meanwhile, the threshold value is adaptively and dynamically adjusted under different working conditions, the false positive rate of the strong turbulent flow condition is reduced, and accurate decision support is provided for operation and maintenance personnel.

A circuit and method of use thereof is provided, the circuit comprising: a first intermediate memory communicatively coupled with a vector processor and a RAM, wherein the vector processor is communicatively coupled with the RAM; an address sequence memory to store non-linear RAM addresses corresponding to linear positions in the first intermediate memory; a data sequencer to read a first frame of data from the RAM to the first intermediate memory based on addresses stored in the address sequence memory; and the first intermediate memory to provide the linearized frame of data to the vector processor to execute a vector instruction.