IEEE SLT 2021 Alpha-mini Speech Challenge: Open Datasets, Tracks, Rules\n and Baselines

摘要

The IEEE Spoken Language Technology Workshop (SLT) 2021 Alpha-mini Speech\nChallenge (ASC) is intended to improve research on keyword spotting (KWS) and\nsound source location (SSL) on humanoid robots. Many publications report\nsignificant improvements in deep learning based KWS and SSL on open source\ndatasets in recent years. For deep learning model training, it is necessary to\nexpand the data coverage to improve the robustness of model. Thus, simulating\nmulti-channel noisy and reverberant data from single-channel speech, noise,\necho and room impulsive response (RIR) is widely adopted. However, this\napproach may generate mismatch between simulated data and recorded data in real\napplication scenarios, especially echo data. In this challenge, we open source\na sizable speech, keyword, echo and noise corpus for promoting data-driven\nmethods, particularly deep-learning approaches on KWS and SSL. We also choose\nAlpha-mini, a humanoid robot produced by UBTECH equipped with a built-in\nfour-microphone array on its head, to record development and evaluation sets\nunder the actual Alpha-mini robot application scenario, including noise as well\nas echo and mechanical noise generated by the robot itself for model\nevaluation. Furthermore, we illustrate the rules, evaluation methods and\nbaselines for researchers to quickly assess their achievements and optimize\ntheir models.\n