摘要
本方法利用人工智能语音识别平台,通过对标准语音的还原,实现文本对比,模拟了人类大脑对自然语音的识别及判断过程,达到对语音质量评估的目的.该方法能有效规避现有人工听音、后台抓取分组、测试MOS值打分3种传统方法,因样本集中度干扰、物理音频指标表征不明确,造成的与用户感知相差较远、效率低等问题;实现了从音频等物理信号对比到语音语义对比的转变,并将人工智能、语音识别等因素加入语音质量评估过程中,切实解决传统方法打分冰冷,无法模拟人脑思维模式的难题.
Artificial intelligence speech recognition platform was used to realize text comparison by restofing standard speech,the recognition and judgment process of natural speech WaS simulated by human brain,and the purpose of speech quality evaluation was achieved.This method can effectively avoid the existing three traditional methods of artificial listening,background grabbing grouping and test MOS score.Because of the interference of sample concentration and the unclear representation of physical audio index,problems such aS far from user perception and low efficiency were caused.The transformation from physical signal comparison such as audio to speech semantic comparison was realized.Artificial intelligence,speech recognition and other factors were added to the process of speech quality aSsessment to effectively solve the problem that traditional methods can’t simulate human brain thinking mode.
作者
周徐
方东旭
文冰松
成昊
ZHOU Xu;FANG Dong-xu;WEN Bing-song;CHENG Hao(China Mobile Group Chongqing Co.,Ltd.,Chongqing 400021,China)
出处
《电信工程技术与标准化》
2018年第A01期48-52,共5页
Telecom Engineering Technics and Standardization
关键词
人工智能
语音识别
语音质量评估
artificial intelligence
speech recognition
speech quality assessment
