The Fifth 'CHiME' Speech Separation and Recognition Challenge: Dataset, Task and Baselines

The CHiME challenge series aims to advance robust automatic speech (ASR) technology by promoting research at the interface of speech language processing, signal processing, and machine learning. This paper the 5th CHiME Challenge, which considers the task of distant-microphone conversational ASR in real home environments. Speech material elicited using a dinner party scenario with efforts taken to capture data is representative of natural conversational speech and recorded by 6 microphone arrays and 4 binaural microphone pairs. The challenge a single-array track and a multiple-array track and, for each track, rankings will be produced for systems focusing on robustness with to distant-microphone capture vs. systems attempting to address all of the task including conversational language modeling. We discuss the for the challenge and provide a detailed description of the data procedure, the task, and the baseline systems for array, speech enhancement, and conventional and end-to-end ASR.