首页> 外文学位 >Improved hidden Markov model partial tracking for additive synthesis using time-frequency analysis.
【24h】

Improved hidden Markov model partial tracking for additive synthesis using time-frequency analysis.

机译:改进的隐马尔可夫模型局部跟踪,用于使用时频分析进行加法合成。

获取原文
获取原文并翻译 | 示例

摘要

Additive synthesis models are popular for musical applications because they offer precise control over the temporal-spectral evolution of sound. The main difficulty in additive modelling lies in the estimation and tracking of model parameters for sounds with time-varying frequency content (this is especially common in music). In the work presented here, a modification is proposed to the hidden Markov model developed by Depalle et al. which improves the estimation and tracking of partial frequencies for additive synthesis. The Wigner-Ville distribution is used in conjunction with the Hough transform in order to re-estimate the frequency and chirp rate from the short-term spectrum. These estimates are then used to formulate a new objective function to score state transitions in the hidden Markov model. The system performance is evaluated using a suite of real and synthetic test signals, and compared to state-of-the-art analysis tools. A clear improvement is demonstrated for highly non-stationary monophonic and polyphonic sounds. Additionally, crossing partials are detected to a limited extent, although further work is still needed to address this problem. This work identifies many strategies to improve additive modelling and suggests several directions for future research.
机译:加法合成模型在音乐应用中很受欢迎,因为它们可以精确控制声音的时频谱演化。加性建模的主要困难在于具有随时间变化的频率内容的声音的模型参数的估计和跟踪(这在音乐中尤其常见)。在这里提出的工作中,提出了对Depalle等人开发的隐马尔可夫模型的修改。这改善了加法合成的部分频率的估计和跟踪。 Wigner-Ville分布与Hough变换一起使用,以便从短期频谱中重新估计频率和线性调频率。然后,将这些估计值用于制定新的目标函数,以对隐马尔可夫模型中的状态转换进行评分。使用一组真实和合成的测试信号评估系统性能,并将其与最新的分析工具进行比较。对于高度非平稳的单声道和多声道声音,已显示出明显的改进。另外,尽管仍然需要进一步的工作来解决这个问题,但是在有限的程度上检测到交叉的部分。这项工作确定了改善加性建模的许多策略,并为未来的研究提出了一些方向。

著录项

  • 作者

    Kereliuk, Corey Mose.;

  • 作者单位

    McGill University (Canada).;

  • 授予单位 McGill University (Canada).;
  • 学科 Music.;Physics Acoustics.;Engineering Electronics and Electrical.
  • 学位 M.A.
  • 年度 2008
  • 页码 92 p.
  • 总页数 92
  • 原文格式 PDF
  • 正文语种 eng
  • 中图分类
  • 关键词

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号