Acoustic Modeling for Overlapping Speech Recognition: Jhu Chime-5 Challenge System

By Vimal Manohar, Szu-Jui Chen, Zhiqi Wang, Yusuke Fujita, Shinji Watanabe, Sanjeev Khudanpur. 2019

21 citations recorded by the source corpus.

This paper summarizes our acoustic modeling efforts in the Johns Hopkins University speech recognition system for the CHiME-5 challenge to recognize highly-overlapped dinner party speech recorded by multiple microphone arrays. We explore data augmentation approaches, neural network architectures, front-end speech dereverberation, beamforming and robust i-vector extraction with comparisons of our in-house implementations and publicly available tools. We finally achieved a word error rate of 69.4% on the development set, which is a 11.7% absolute improvement over the previous baseline of 81.1%, and release this improved baseline with refined techniques/tools as an advanced CHiME-5 recipe.

Subfields: Artificial Intelligence, Speech & Audio, Signal Processing

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

1. Reinhold Häb-Umbach
2. Catalin Zorila
3. Piotr Żelasko
4. Christoph Boeddeker
5. Rama Doddipatla
6. Yishay Carmiel

In Speech & Audio

1. Reinhold Häb-Umbach
2. Piotr Żelasko
3. Christoph Boeddeker
4. Rama Doddipatla

In Signal Processing

1. Reinhold Häb-Umbach
2. Catalin Zorila
3. Piotr Żelasko
4. Christoph Boeddeker
5. Rama Doddipatla
6. Yishay Carmiel
7. Jinyu Li
8. Shinji Watanabe

Citations in the corpus

• DiPCo - Dinner Party Corpus (2019, unknown): “Baseline word error rate (WER) numbers were generated by using the Kaldi baseline (egs/chime5/s5) [2, 3] as the acoustic model (AM) source on which model adaptation was performed.”
• An Investigation into the Effectiveness of Enhancement in ASR Training and Test for Chime-5 Dinner Party Transcription (2019, methodology): “All TDNN-F layers have the topology described above.”
• WER we are and WER we think we are (2020, background): “On the other hand, on the truly challenging benchmarks, such as the dinner party conversations CHiME5 (Manohar et al., 2019) benchmark, modern ASRs report WERs in the ranges of 46%–73%.”