Front-end Signal Processing for Improved Automatic Speech Recognition

用于改进自动语音识别的前端信号处理

• 批准号: 522072-2017
• 负责人: Saucier, Antoine
• 金额: $ 1.82万
• 依托单位: École Polytechnique de Montréal
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2017
• 资助国家: 加拿大
• 起止时间: 2017-01-01 至 2018-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=639626
• 关键词: Front; end; Signal; Processing; Improved

Our partner Fluent.AI develops Automatic Speech Recognition (ASR) systems based on neural networks. Theperformance of ASR systems degrades significantly in the presence of noise. The types of noises that arerelevant occur e.g. on urban sidewalks, construction sites, factory floors or in cars (e.g. radio music orbackground conversations). The performance degradation of ASR systems is partly caused by a lack ofrobustness to noise of the data representations used in ASR. To improve the performance of our partner's ASRsystems, we propose to develop a pre-processing method that reduces noises in the speech signal before it issent to the ASR engine. Our method is based on independent component analysis.

我们的合作伙伴Fluent.ai开发基于神经网络的自动语音识别（ASR）系统。在存在噪声的情况下，ASR系统的性能大大降低。 Arerelevant发生的噪声类型，例如在城市人行道，建筑工地，工厂地板或汽车上（例如广播音乐或背景对话）。 ASR系统的性能下降部分是由于缺乏对ASR中使用的数据表示的噪声的缺乏。为了提高合作伙伴的ASRSystems的性能，我们建议开发一种预处理方法，该方法在对ASR发动机发出之前会减少语音信号中的声音。我们的方法基于独立的组件分析。