arXiv subjects

Kun Wei

Publications and source records attributed to Kun Wei.

39 records · Page 3Linked to original sources

arXiv2020

Context-aware RNNLM Rescoring for Conversational Speech Recognition

Conversational speech recognition is regarded as a challenging task due to its free-style speaking and long-term contextual dependencies. Prior work has explored the modeling of long-range context through RNNLM rescoring with improved performance. To further take advantage of the persisted nature during a conversation, such as topics or speaker turn, we extend the rescoring procedure to a new context-aware manner. For RNNLM training, we capture the contextual dependencies by concatenating adjacent sentences with various tag words, such as speaker or intention information. For lattice rescoring, the lattice of adjacent sentences are also connected with the first-pass decoded result by tag words. Besides, we also adopt a selective concatenation strategy based on tf-idf, making the best use of contextual similarity to improve transcription performance. Results on four different conversation test sets show that our approach yields up to 13.1% and 6% relative char-error-rate (CER) reduction compared with 1st-pass decoding and common lattice-rescoring, respectively.

cs.SD↗

arXiv2020

Recent Developments on ESPnet Toolkit Boosted by Conformer

In this study, we present recent developments on ESPnet: End-to-End Speech Processing toolkit, which mainly involves a recently proposed architecture called Conformer, Convolution-augmented Transformer. This paper shows the results for a wide range of end-to-end speech processing applications, such as automatic speech recognition (ASR), speech translations (ST), speech separation (SS) and text-to-speech (TTS). Our experiments reveal various training tips and significant performance benefits obtained with the Conformer on different tasks. These results are competitive or even outperform the current state-of-art Transformer models. We are preparing to release all-in-one recipes using open source and publicly available corpora for all the above tasks with pre-trained models. Our aim for this work is to contribute to our research community by reducing the burden of preparing state-of-the-art research environments usually requiring high resources.

eess.AS↗

arXiv2003

Global Platform for Rich Media Conferencing and Collaboration

The Virtual Rooms Videoconferencing Service (VRVS) provides a worldwide videoconferencing service and collaborative environment to the research and education communities. This system provides a low cost, bandwidth-efficient, extensible means for videoconferencing and remote collaboration over networks within the High Energy and Nuclear Physics communities (HENP). VRVS has become a standard part of the toolset used daily by a large sector of HENP, and it is used increasingly for other DoE/NSF-supported programs. The current features included multi-protocol, multi-OS support for all significant video enabled clients including: H.323, Mbone, QuickTime, MPEG2, Java Media Framework, and other clients. The current architecture makes VRVS a distributed, highly functional, and efficient software-only system for multipoint audio, video and web conferencing and collaboration over global IP networks. VRVS has developed the VRVS-AG Reflector and a specialized Web interface that enables end users to connect to any Access Grid (AG) session, in any of the AG "virtual venues" from anywhere worldwide. The VRVS system has now been running for the last five and half years, offering to the HENP community a working and reliable tool for collaboration within groups and among physicists dispersed world-wide. The goal of this ongoing effort is to develop the next generation collaborative systems running over next generation networks. The new developments area integrate emerging standards, include all security aspects, and will extend the range of VRVS video technologies supported to cover the latest high end standards quality. We will focus the discussion on the new capability provides by the latest version V3.0 and its future evolution.

cs.MM↗