5 results about "Function word" patented technology

The invention discloses an ancient modern text machine translation method based on deep learning, and belongs to the technical field of natural language processing. The invention provides a multi-task collaborative optimization framework aiming at the problems of domain knowledge deficiency, frequent word activation, corpus scarcity and the like of an existing neural network model in ancient language translation. Precise domain knowledge injection is realized by constructing a closed ancient language parallel corpus retrieval library and combining historical term alignment and logic chain reasoning; a probability weighted part-of-speech embedding mechanism is adopted, and grammar feature distribution of virtual word ambiguity and word type activity is dynamically fused, so that the grammar analysis robustness is improved; and semantic generation, retrieval alignment and part-of-speech constraint tasks are jointly optimized through a fixed weight strategy. According to the method, the semantic accuracy and the logic continuity of ancient language translation are improved, and extended application of complex ancient language styles is supported. Experimental results verify that the performance of the method is improved in terms of BLEU and ROUGE-L indexes, and efficient technical support is provided for ancient book digitization and cross-time explanation.

This invention discloses a text pronunciation optimization method and system for speech synthesis, belonging to the field of speech synthesis management technology. In this method, after emotional processing, long sentences are split into segments based on a triple rule system of semantic blocks, classical Chinese function words, and character length. The English portion of the text undergoes layered processing, distinguishing between pure uppercase letter combinations and regular English words, adding splitting and prosodic markers to pure uppercase letter combinations, and generating optimized text pronunciation information. This invention breaks through the limitations of existing single-rule adaptation in speech synthesis text processing, pioneering a multi-dimensional layered optimization framework that integrates technologies such as semantic parsing of numbers and operators, scene-based matching of polyphonic characters, emotional markers for literary texts, and semantic long sentence splitting. It designs multiple exclusive optimization rules for the speech synthesis and reading needs of literary and everyday spoken texts, solving the core pain point of existing technologies that emphasize generality but neglect specific scenarios.

The invention relates to an efficacy word extraction system and method based on multi-modal adaptive learning. According to the system, multi-modal data such as texts, images and audios are received through the input module, and feature fusion is carried out through the preprocessing module. A hybrid encoder module is utilized, local n-gram features are firstly extracted through a convolutional neural network layer, then global context information is extracted through a Transform encoder, the comprehensiveness of feature representation is guaranteed through a cooperation mechanism, and the limitation of a traditional single model in the aspect of feature representation precision is overcome. And then, the knowledge distillation module optimizes the text semantic representation data to generate efficient lightweight semantic representation information. The decoder module integrates a vocabulary library, adopts a strategy of combining generation and copying, and utilizes the thought of a pointer-generation network, so that the problem of unregistered words is effectively solved, the coverage rate and accuracy are ensured while generalization is kept, and efficient, accurate and robust efficacy word extraction is realized.

This invention belongs to the field of image and text processing, and discloses a visual language model training method, image and text analysis method, and related apparatus. It obtains the function word mask of text samples based on a pre-set function word dictionary; utilizes the visual encoder and text encoder of the visual language model to obtain the corresponding image features and complete text features of image and text samples, as well as the function word text features obtained from the aforementioned function word mask; calculates the text-image cross-attention and function word-image cross-attention using the aforementioned image features, complete text features, and function word text features, and then obtains the final differential cross-attention through difference; then calculates the training loss and updates the parameters through backpropagation gradients; by repeating the above steps, the visual language model is trained, resulting in a visual language model with low training cost, almost lossless accuracy, and improved robustness, thereby enhancing the image and text analysis capabilities of the visual language model and providing an optimized foundation for the security protection of the visual language model system.

The invention discloses a Tibetan rhythm structure prediction method based on grammar information, relates to the technical field of speech synthesis, and is applied to the field of rhythm structure prediction in Tibetan speech synthesis. The syntactic information-based Tibetan rhythm structure prediction method comprises the following steps of S1, completing Tibetan analysis and virtual word continuous quantization, extracting a grammar probability, a hierarchical focus and a pause difference, and storing and constructing a Tibetan rhythm prediction database after preprocessing; s2, based on the strength of the virtual word continuing relation, the influence range hit mark and the word order distance, boundary discrimination is carried out; s3, entropy feature analysis is carried out through the grammar role probability data and the boundary judgment data; s4, generating a three-layer rhythm boundary type sequence according to boundary forming, and inhibiting over-dense pause; and S5, carrying out rhythm intensity evaluation through the pause duration difference and the grammar information entropy data. The problems of pause, confusion and increased understanding burden caused by dislocation of rhythm and semantic levels in Tibetan long speech synthesis are solved.