| In recent years,the traditional color TV industry has been developing intelligently,and with the continuous improvement of artificial intelligence,it has become an important direction for the development of smart TV through artificial intelligence technology to improve the intelligence level of television.It is in such an environment that functions such as generating a summary of a text content displayed by a smart TV and broadcasting it by voice are proposed.This article mainly describes the design and development of smart TV text content summary generation and voice broadcast function.First,perform technical research and analysis on abstract generation and speech synthesis as core functions,determine the direction of function implementation based on deep learning,and then compare and analyze existing deep learning-based abstract generation and speech synthesis technologies.The advantages and disadvantages of various algorithm technologies,a new two-way summary generation model based on transfer learning and a speech synthesis model based on Tacotron2 are proposed.After giving a comprehensive algorithm scheme,each function is detailed In the experiment,through the reproduction of the existing mature algorithm model and the simulation experiment and experimental analysis of the algorithm model proposed in this paper,the superiority of the proposed algorithm scheme was determined,and the method based on transfer learning was determined.The two-way summary generation model and Tacotron2 speech synthesis model are used as technical support to realize the core functions of the system.Then detailed requirements analysis and modeling of the system development are carried out.Through the requirements analysis of the system,the system is identified as background management,summary generation,Speech synthesis and model compression four models,and given The example diagram determines the function of each module,and then according to the complete data flow diagram,the business use for administrators and ordinary users is described in detail according to different processes;through detailed analysis of the module,the detailed design of each module is given And implement it;at last,according to the requirements analysis and detailed design of each module,detailed functional tests and performance tests were performed on each module to ensure the effectiveness and performance guarantee of each functional module of the system and to complete the system Run the test to ensure the smooth operation of the system.The system innovation points proposed in this article are: from the functional perspective,a complete solution for generating smart TV text content and voice announcements to users is provided.Through technical integration,the convenience of users using smart TVs is greatly improved.Improved user experience;technically,taking deep learning technology as the starting point,a bidirectional language model that is very effective for this topic is proposed as a feature extractor,whose performance matches that of the BERT language model,and at the same time for the existing abstract generated The Seq2 Seq architecture is improved,and a bidirectional digest generation decoder is designed and implemented.The bidirectional language model is used to fine-tune the digest generation task,which greatly improves the task performance.Finally,we use the model compression technology based on weight pruning to deepen the depth involved in this article.Learning models for processing,accelerating model inference,reducing hardware requirements and facilitating the need for projects to land in the future.As a pre-research project at an internship company,this function was successfully developed and passed the company's acceptance,which confirms that the improvement of the intelligent level of smart TVs is the future trend,and the integration of technologies in multiple fields will improve the intelligence.The huge role that artificial intelligence technology plays in it is also immeasurable. |
PDF Full Text Request