ESPnet-se: end-to-end speech enhancement and separation toolkit designed\n for asr integration

Abstract

We present ESPnet-SE, which is designed for the quick development of speech\nenhancement and speech separation systems in a single framework, along with the\noptional downstream speech recognition module. ESPnet-SE is a new project which\nintegrates rich automatic speech recognition related models, resources and\nsystems to support and validate the proposed front-end implementation (i.e.\nspeech enhancement and separation).It is capable of processing both\nsingle-channel and multi-channel data, with various functionalities including\ndereverberation, denoising and source separation. We provide all-in-one recipes\nincluding data pre-processing, feature extraction, training and evaluation\npipelines for a wide range of benchmark datasets. This paper describes the\ndesign of the toolkit, several important functionalities, especially the speech\nrecognition integration, which differentiates ESPnet-SE from other open source\ntoolkits, and experimental results with major benchmark datasets.\n