In this paper, we present an efficient look-ahead technique based on both the Language Model (LM) Look-Ahead and the Acoustic Model (AM) Look-Ahead, for the time-synchronous beam search in the large vocabulary speech ...
详细信息
ISBN:
(纸本)7801501144
In this paper, we present an efficient look-ahead technique based on both the Language Model (LM) Look-Ahead and the Acoustic Model (AM) Look-Ahead, for the time-synchronous beam search in the large vocabulary speech recognition. In this so-call stage based look-ahead (SLA) technique, two predicting processes with different hypothesis evaluating criteria are organized by stages according to the different requirements for pruning the unlikely surviving hypotheses. Furthermore, in order to reduce the efforts for distributing the LM over the lexical tree more effectively, the LM Rank based Pruning (LMRP) is integrated with the extension of each new phoneme node. The recognition experiments performed on the 50k-word Mandarin Dictation task (Easytalk2000) show that a reduction by 10 percents in the search effort in comparison with the standard word-conditioned search using LM look-ahead only, and a reduction of 25 percents in the word error rates in comparison with the search algorithm without any look-ahead can be achieved.
When using neural networks to train a large number of data for classification, there generally exists a learning complexity problem. In this paper, a new geometrical interpretation of McCulloch-Pitts (M-P) neural mode...
详细信息
Real-time path planning is a key subject in the research of a Multi-robot system, such as in soccer robot system, as in such system there are several robots to coordinate to complete one task and its workspace is time...
详细信息
Real-time path planning is a key subject in the research of a Multi-robot system, such as in soccer robot system, as in such system there are several robots to coordinate to complete one task and its workspace is time-varying. There exist unpredictable changes in such workspace. These changes will make the robots path planning tremendously hard. This paper introduce the study on the real-time planning of a multi-robot system, such as soccer robots system. In such systems, as there exist multiple mobile robots, robots path planning is highly required of real-time and of high path planning quality. In this paper, the definitions of Active Obstacle and Dominating Robot are put forward. Genetic-planning and Fuzzy-correction are developed as our path planning techniques. The concept of Sector Vital Value is proposed for Genetic-planning and Fuzzy-correction. Block-Inserting strategy is employed to further improve the system's real-time performance in the workspace modelization. It takes only 2% of workspace modeling time required by Grid-modeling technique. After taking the proposed measures, the system's real-time performance is improved a lot. The initial simulation shows that our planning strategy can meet real-time demand.
A neural network(NN)-based adaptive controller with an observer is proposed in this paper for trajectory tracking of robotic manipulators with unknown dynamic nonlinearities. It is assumed that the robotic manipulator...
详细信息
A neural network(NN)-based adaptive controller with an observer is proposed in this paper for trajectory tracking of robotic manipulators with unknown dynamic nonlinearities. It is assumed that the robotic manipulator has only joint angle position measurements. The NN is used to approximate the revised robot dynamics function. The adaptive controller with an observer can guarantee the uniform ultimate bounds of the tracking errors and the observer errors as well as the bounds of the NN weights. Main theoretical results for designing such a controller are given, and the control performance of the proposed controller is verified with simulation studies.
The paper puts forward a quasidependency model for structural analysis of Chinese i baseNPs and a MDL-based algorithm for quasidependency-strength acquisition. The experiments show that the proposed model is more suit...
详细信息
This paper describes a specific VLSI design of fuzzy micro-controller with advanced pipeline *** architecture has the distinctive advantage of using single instruction-flow and multi-dataflow and adopting some advance...
详细信息
This paper describes a specific VLSI design of fuzzy micro-controller with advanced pipeline *** architecture has the distinctive advantage of using single instruction-flow and multi-dataflow and adopting some advanced structures including: superscalar pipeline,centrally dynamic scheduling of pipeline and precontract table,*** to these technologies the circuit has satisfactory performances, clock cycles per fuzzy instruction(CPI) is *** circuit is simulated with very fast hardware description language (VHDL) and synthesis results show that if implemented with 1.5 urn CMOS technology lib,it will cover 20K unit area (include 2K RAM),the average speed of data processing reach six million rules per second and highest clock cycle frequency reach thirty million *** also compare our design with Neuralogix Incorporation s'commercially available NLX230.
This paper proposes an approach for speech recognition using recurrent neural networks(RNN) and corresponding training schema,in which RNN is trained as a digit *** order to reduce the training cost and improve the pe...
详细信息
This paper proposes an approach for speech recognition using recurrent neural networks(RNN) and corresponding training schema,in which RNN is trained as a digit *** order to reduce the training cost and improve the performance,several techniques had been tried in our *** weights of the net are trained by stages and the initial feedback values of RNN are trained along with the *** result shows that these methods improve the recognition accuracy by 10%.
Hardware description language (HDL) code designing is a critical component of the chip design process, requiring substantial engineering and time resources. Recent advancements in large language models (LLMs), such as...
详细信息
Hardware description language (HDL) code designing is a critical component of the chip design process, requiring substantial engineering and time resources. Recent advancements in large language models (LLMs), such as GPT series, have shown promise in automating HDL code generation. However, current LLM-based approaches face significant challenges in meeting real-world hardware design requirements, particularly in handling complex designs and ensuring code correctness. Our evaluations reveal that the functional correctness rate of LLM-generated HDL code significantly decreases as design complexity increases. In this paper, we propose the AutoSilicon framework, which aims to scale up the hardware design capability of LLMs. AutoSilicon incorporates an agent system, which 1) allows for the decomposition of large-scale, complex code design tasks into smaller, simpler tasks; 2) provides a compilation and simulation environment that enables LLMs to compile and test each piece of code it generates; and 3) introduces a series of optimization strategies. Experimental results demonstrate that AutoSilicon can scale hardware designs to projects with code equivalent to over 10,000 tokens. In terms of design quality, it further improves the syntax correctness rate and functional correctness rate compared with approaches that do not employ any extensions. For example, compared to directly generating HDL code using GPT-4-turbo, AutoSilicon enhances the syntax correctness rate by an average of 35.8% and improves functional correctness by an average of 35.6%.
A two-stage electroencephalogram (EEG) source tomography method is proposed. First, the layer recursive algorithm (LRA) is developed to calculate the cortical potentials from the scalp potentials by using the finite v...
详细信息
暂无评论