面向多场景的环境异常声音识别

随着人工智能和大数据的发展，各种场景中对异常声音识别的需求日益增长，基于人工智能的声音识别技术正在兴起并被高度重视。现行主流的异常声音识别算法多为浅层机器学习模型结构，对异常声音的识别率较低，且识别的声音类型单一。为了有效识别异常声音，提出一种基于梅尔频率倒谱系数(Mel-frequency cepstral coefficient, MFCC)和卷积神经网络(convolution neural network, CNN)的环境声音识别算法，对各类异常声音进行采集和有效识别，并及时反馈声音状态，为各类声识别应用场景提供精细化管理技术手段。结果表明：提出的算法对5类场景下环境异常声音的识别率得到极大提高，适用于更广泛的声学场景，具有明显的优势。

Environmental Abnormal Sound Recognition for Multiple Scenes

With the development of artificial intelligence and big data, there is an increasing demand for abnormal sound recognition in various scenes. Sound recognition technology based on artificial intelligence is emerging and is highly valued. Most of the current mainstream abnormal sound recognition algorithms are shallow machine learning model structure. The recognition rate of abnormal sound is low, and the recognition of the sound type is single. In this paper, a high effective algorithm of sound recognition based on Mel-frequency cepstral coefficient (MFCC) and Convolution neural network (CNN) is proposed. It could collect all kinds of abnormal sounds and feed back the sound state in time, which provides effective management technical means for all kinds of sound recognition application scenarios. The results show that the recognition rate of the proposed algorithm for five kinds of abnormal sounds in the environment is greatly improved, which is suitable for a wider range of acoustic scenes and has obvious advantages.