• Phonetics and Speech Sciences
• /
• v.12 no.2
• /
• pp.29-37
• /
• 2020

Recently, the neural network-based deep learning algorithm has dramatically improved performance compared to the classical Gaussian mixture model based hidden Markov model (GMM-HMM) automatic speech recognition (ASR) system. In addition, researches on end-to-end (E2E) speech recognition systems integrating language modeling and decoding processes have been actively conducted to better utilize the advantages of deep learning techniques. In general, E2E ASR systems consist of multiple layers of encoder-decoder structure with attention. Therefore, E2E ASR systems require data with a large amount of speech-text paired data in order to achieve good performance. Obtaining speech-text paired data requires a lot of human labor and time, and is a high barrier to building E2E ASR system. Therefore, there are previous studies that improve the performance of E2E ASR system using relatively small amount of speech-text paired data, but most studies have been conducted by using only speech-only data or text-only data. In this study, we proposed a semi-supervised training method that enables E2E ASR system to perform well in corpus in different domains by using both speech or text only data. The proposed method works effectively by adapting to different domains, showing good performance in the target domain and not degrading much in the source domain.

• The KIPS Transactions:PartC
• /
• v.13C no.7 s.110
• /
• pp.803-812
• /
• 2006

This paper analytically evaluates the FEC(Forward Error Correction) symbol size's effect on the performance and energy consumption of 802.11 protocol with the block FEC algorithm over WSN(Wireless Sensor Network). Since the basic recovery unit of block FEC algorithms is symbols not bits, the FEC symbol size affects the packet correction rate even with the same amount of FEC check bits over a given WSN channel. Precisely, when the same amount of FEC check bits are allocated, the small-size symbols are effective over channels with frequent short bursts of propagation errors while the large ones are good at remedying the long rare bursts. To estimate the effect of the FEC symbol site, the paper at first models the WSN channel with Gilbert model based on real packet traces collected over TIP50CM sensor nodes and measures the energy consumed for encoding and decoding the RS (Reed-Solomon) code with various symbol sizes. Based on the WSN channel model and each RS code's energy expenditure, it analytically calculates the transmission efficiency and power consumption of 802.11 equipped with RS code. The computational analysis combined with real experimental data shows that the RS symbol size makes a difference of up to 4.2% in the transmission efficiency and 35% in energy consumption even with the same amount of FEC check bits.

• Journal of KIISE:Information Networking
• /
• v.36 no.5
• /
• pp.425-436
• /
• 2009

In recent years, the user demands have increased for multimedia service of high quality over the broadband convergence network. These rising demands for high quality multimedia service led the popularization of various user terminals and large scale display equipments, which needs a variety type of QoS (Quality of Service). In order to support demands for QoS, numerous research projects are in progress both from the perspective of network as well as end system; For example, at the network perspective, QoS guaranteeing by improving of internet performance such as Active Queue Management, while at the end system perspective, SVC (Scalable Video Coding) encoding scheme to guarantee media quality. However, existing AQM algorithms have problems which do not guarantee QoS, because they did not consider the essential characteristics of video encoding schemes. In this paper, it is proposed to solve this problem by deploying the TS- AQM (Temporal Scalability Active Queue Management) which employs the differentiated packet dropping for dependency of the temporal level among the frames, based on SVC encoding characteristics by exploiting the TID (Temporal ID) field of the SVC NAL unit header. The proposed TS-AQM guarantees multimedia service quality through video decoding reliability for SVC streaming service, by differentiated packet dropping when congestion exists.