汽车内驾驶员语音增强评价研究

Research on speech enhancement evaluation of driver in vehicle

驾驶员语音增强质量的评价指标是保证语音增强算法性能的关键,而现有的语音增强质量评价方法不能准确地反映人对声音感知的主观性。针对上述问题,分析了言语可懂度指数对语音增强算法评价的适用性,并在某品牌汽车上进行实验。通过在汽车内建立均匀线性传声器阵列来对驾驶员语音进行信号采集,然后利用波束形成算法对阵列中不同传声器组合的语音信号进行增强,得到汽车在不同行驶速度时不同阵列组合的语音增强结果。使用信噪比和言语可懂度指数分别对实验得到的语音增强结果进行评价,结果表明言语可懂度指数更适合评价汽车内驾驶员的语音增强算法的性能。

The evaluation index of speech enhancement quality for drivers is the key to ensure the performance of speech enhancement algorithm,but the existing methods for speech enhancement quality evaluation cannot accurately reflect the subjectivity of human sound perception.To solve this problem,the applicability of speech intelligibility index to the evaluation of speech enhancement algorithm is analyzed,and experiments on a certain brand of vehicle are conducted.The driver's voice signal is collected by building a uniform linear microphone array in the vehicle,and then the beamforming algorithm is used to enhance the voice signals of different microphone combinations in the array.The speech enhancement effects of different microphone combinations at different driving speeds are obtained and evaluated by using signal-to-noise ratio(SNR)and speech intelligibility index(SII),and the results show that the speech intelligibility index is more suitable for evaluating the performance of the speech enhancement algorithms.