The third ‘CHiME’ speech separation and recognition challenge: Analysis and outcomes

• Published in: Computer speech & language Vol. 46; pp. 605 - 626
• Main Authors: Barker, Jon, Marxer, Ricard, Vincent, Emmanuel, Watanabe, Shinji
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.11.2017; Elsevier
• Subjects: Computer Science; Microphone array; Noise-robust ASR; Signal and Image Processing; ‘CHiME’ challenge; ‘CHiME’ challenge; Microphone array; Noise-robust ASR; 'CHiME' challenge; microphone array
• DOI: 10.1016/j.csl.2016.10.005
• ISSN: 0885-2308

•The presentation of a unique multi-microphone speech recognition challenge with speech recorded in real environments.•A detailed characterisation of the challenge audio using novel analyses to estimate key properties of the speakers, environments and noisy speech signals.•An overview of 26 systems submitted to the challenge presenting a snapshot of the state-of-the-art in distant microphone ASR.•A presentation of system performance identifying which signal processing and statistical modelling techniques are the most beneficial.•A presentation of correlations between signal characteristics and system performances across utterances addressing the question, “What are the particular circumstances that lead to high word error rates?” This paper presents the design and outcomes of the CHiME-3 challenge, the first open speech recognition evaluation designed to target the increasingly relevant multichannel, mobile-device speech recognition scenario. The paper serves two purposes. First, it provides a definitive reference for the challenge, including full descriptions of the task design, data capture and baseline systems along with a description and evaluation of the 26 systems that were submitted. The best systems re-engineered every stage of the baseline resulting in reductions in word error rate from 33.4% to as low as 5.8%. By comparing across systems, techniques that are essential for strong performance are identified. Second, the paper considers the problem of drawing conclusions from evaluations that use speech directly recorded in noisy environments. The degree of challenge presented by the resulting material is hard to control and hard to fully characterise. We attempt to dissect the various ‘axes of difficulty’ by correlating various estimated signal properties with typical system performance on a per session and per utterance basis. We find strong evidence of a dependence on signal-to-noise ratio and channel quality. Systems are less sensitive to variations in the degree of speaker motion. The paper concludes by discussing the outcomes of CHiME-3 in relation to the design of future mobile speech recognition evaluations.