对话、念读两种说话状态下声纹特征变化的实验研究

Experimental study on changes of voiceprint features in dialogue and reading

在对话、念读两种说话状态下,对20位青年男性普通话语流中240个音段所包含的单元音[a]、[i]、[u]的共鸣方式、平均音高、音强等听觉特征,共振峰走向、强度、频率等频谱特征进行定性和定量分析。发现说话状态改变后个人的声纹特征有着不同程度的变化:听觉特征、共振峰强度特征的变化较大,共振峰走向和频率特征相对稳定。在被检测的音段中,40.0%的音段听感音高和音强发生了变化,45.3%的音段共振峰强度发生了变化;尽管有13.3%的音段共振峰走向存在差异,但共振峰频率变异系数的变化范围均在2.7%以内。由此可见,在对不同说话状态下的语音进行声纹鉴定时,应重点关注共振峰走向和频率,以判断涉案语音和嫌疑人样本语音是否同一。

In the two speaking states of dialogue and reading, this paper makes qualitative and quantitative analysis on the resonance mode, average pitch, sound intensity and other auditory characteristics of the unit sound [a], [i], [u] contained in 240 segments in the Mandarin language flow of 20 young men, as well as the spectrum characteristics such as formant trend, formant intensity and formant frequency. In the detected segments, 40.0% of the segments have changed the pitch and intensity of the sound, and 46.0% of the segments have changed their peak intensity. Although 13.3% of the segments have different formants, the variation range of formant frequency variation coefficient is within 2.7%. Therefore, it can be seen that when the voiceprint identification of different speech states is carried out, more attention should be paid to the formant trend and frequency to judge whether the voice involved is the same as that of the suspect sample.