[ Japanese | English ]
Showa-ku, Nagoya 4668555 JAPAN
ri@nitech.ac.jp
Welcome! Here at Lee Laboratory of Nagoya Institute of Technology, we are conducting research on speech recognition, speech dialogue, natural language processing, speech interface, and speech interaction, targeting human-to-human and human-to-machine communication using speech and language.
Our aim is to develop technologies for intelligent information processing, including speech, language, and dialogue, as well as realize natural and easy-to-use voice and language interfaces.
LEE Akinobu was born in Kyoto, Japan, on December 19, 1972. He received the B.E. and M.E. degrees in information science, and the Ph.D. degree in informatics from Kyoto University, Kyoto, Japan, in 1996, 1998 and 2000, respectively. He worked on Nara Institute of Science and
Technology as an assistance professor from 2000-2005. Currently he
is a professor of Nagoya Institute of Technology, Japan. His research interests include speech recognition, spoken language understanding, and spoken dialogue system. He is a member of IEEE, ISCA, JSAI, IPSJ and the Acoustical Society of Japan. He is also a developer of open-source speech recognition software Julius and CG agent-based speech interaction toolkit MMDAgent.
Assistant Professor Sei Ueno joined Lee lab in April. Together with Assist. Prof. Ueno, we will continue our research on spoken language information processing and spoken dialogue interfaces.
From December 2020, Our lab has been participating in the “Avatar Symbiotic Society” project, a moonshot-type research and development project led by Professor Ishiguro of Osaka University.
“CG-Specific Dialogue” is our research theme. We are focusing on a new conversation system with CG characters that seamlessly integrates an autonomous dialogue system and human remote control (avatars) to realize truly usable, rich-enpowered human-communication system on the next era. We are now working in collaboration with various research institutes.
Julius version 4.6 has been released. You can get it from its GitHub site.
What’s new in Julius-4.6 Julius-4.6 is a minor release with new features and fixes, including GPU integration and grammar handling updates.
GPU-based DNN-HMM computation (Take a look at v4.6 performance comparison on YouTube!)
Now Julius can compute DNN-HMM with GPU. Total decoding will be four times faster than CPU-based computation on Julius-4.5.
Requires CUDA version 8, 9 or 10.
Julius has merged a pull request that adds a new feature “grammar search on the 1st pass”. To use it, get the latest code on master branch.
It enables applying full grammar on the 1-pass, thus outputs more reliable (grammar-constrained) result at the 1st pass.
Background The grammar-based recognition on Julius does not apply the full grammar on the 1st pass, but applies only the word-pair constraint extracted from the grammar for efficiency.
The graduation thesis presentation meeting was held. The following ten members gave presentations:
松岡 優太「楽曲の再生履歴情報を用いた GA による自動メロディ生成」 中川 樹「ホルンを対象とした音響信号による音色悪化要因判別」 尾関 日向「音楽音響信号を対象としたギターパートの分離」 池田 将「対話的楽曲推薦を目的とした発話からの感情要素と感情強度推定手法」 高井 幸輝「マルチコーパスおよびマルチラベルを用いた Pre-training Fusion による自発音声の感情分類」 岡本 空大「ニューラルネットワークを用いた講演音声の自動発話印象推定における推定精度の向上」 大平原 海斗「Weakly-labeled Data を用いた音響イベント検出」 加藤 弘泰「音声対話システムにおけるユーザ話速およびコーパスに基づく話速制御」 西山 達也「話者情報を用いたBERTによる複数人対話における応答選択」 石島 侑弥「対話状態追跡における対話行為タグを用いた重要対話履歴抽出」