The USTC-iFlytek systems for CHiME-5 Challenge
By Jun Du, Tian Gao, Lei Sun, Feng Ma, Yi Fang, Diyuan Liu, Q. Zhang, Xiang Zhang, Haikun Wang, Jia Pan and 3 more. 2018
27 citations recorded by the source corpus.
This report describes our submission to the fifth CHiME Challenge. The main technical points of our system include the deep learning based speech enhancement and separation, training data augmentation via different versions of the official training data, SNR-based array selection, front-end model fusion, acoustic model fusion, and language model fusion. Tested on the development test set, our best system for single-array track using official LM has yielded a 37.7% WER relative reduction over the results given by official baseline system.
Subfields: Artificial Intelligence, Computer Networks and Communications, Hardware and Architecture, Speech & Audio
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Shinji Watanabe
- Shota Horiguchi
- Yusuke Fujita
- Naoyuki Kanda
- Kenji Nagamatsu
- Jinyu Li
- Takaaki Hori
- Yawen Xue
In Speech & Audio
- Shinji Watanabe
- Yusuke Fujita
- Shota Horiguchi
- Naoyuki Kanda
- Kenji Nagamatsu
- Jinyu Li
- Hervé Bredin
- John R. Hershey
Citations in the corpus
- CHiME-6 Challenge: Tackling Multispeaker Speech Recognition for Unsegmented Recordings (2020, neutral): “, we purposefully did not use system combination but the result is still close to [49], which is based on complex multi-path enhancement processing with system combination.”
- End-to-End Neural Speaker Diarization with Permutation-Free Objectives (2019, background): “It also helps automatic speech recognition performance in multi-speaker conversation scenarios in meetings (ICSI [3,4], AMI [5,6]) and home environments (CHiME5 [6–10]).”