通过整合声门脉冲数据，实现清晰可视化与高噪声下稳定分析。
Non-locally averaged pruned reassigned spectrograms: a tool for glottal pulse visualization and analysis
- 将大量声门脉冲叠加、求和并剪枝，生成简化谱图
- 在高噪声环境下比传统LPC方法更稳定可靠
- 适合语音学研究者分析声带振动模式
重新分配谱图在精确测量共振峰和区分说话人方面表现优异，但难以以直观可复现的方式呈现大量数据。本文基于Fulop和Fitz的方法，提出非局部平均剪枝重新分配谱图（NAPReS），通过堆叠、求和和剪枝大量声门脉冲，清晰展现发音时元音中心处的声门脉动特征。该方法使低幅度周期性结构更易观测，同时支持如高斯混合模型（GMM）等替代共振峰拟合方法。实验表明，在高噪声条件下，使用GMM的NAPReS比传统LPC方法更具可重复性。
原文摘要 · Abstract (English)
Reassigned spectrograms have shown advantages in precise formant measuring and inter-speaker differentiation. However, reassigned spectrograms suffer from their inability to visualize larger amounts of data in an easily comprehensible and reproducible manner. Utilizing the techniques and tools developed by Fulop and Fitz, a variation of the reassigned spectrogram is proposed. Non-locally Averaged Pruned Reassigned Spectrograms (NAPReS) provide a simplified view into the characteristics of a speaker's glottal pulsation patterns throughout the centroid of a vowel through the stacking, summing, and pruning of large numbers of glottal pulses. In this exploratory study, NAPReS has been shown to display a large amount of data in an easily comprehensible and quantifiable manner, while also making the observation of low-amplitude cyclical structures more accessible. NAPReS also allows for alternative formant fitting methods such as Gaussian mixture modeling. In this study, NAPReS with GMM was compared against conventional LPC fitting of formant values and was shown to be more reproducible than conventional LPC fitting in high-noise situations.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。