Affiliations: [a]
School of Music and Dance of Yunnan Normal University, Kunming, China
| [b]
Tarrant County College, TX, USA
Correspondence:
[*]
Corresponding author. Xin Li, School of Music and Dance of Yunnan Normal University, Kunming, China. E-mail: 18669009935@163.com.
Abstract: In this subject, the model recognition method was adopted, namely, the multi-note model based on the hidden Markov process was established by using the multi-note as the basic modeling unit. And the related modules in HTK were recompiled to build a multi-note recognition model, thus the features of a multi-note audio file were extracted; An optimization and updating scheme was proposed from the algorithm flow and evaluation model. By effectively evaluating the data state of the piano note model, an effective recognition model was established, and the corresponding recognition results were given. Then based on the analysis of the principle of commonly used audio file parameterization and combined with the characteristics of multi-note audio, the existing feature extraction modules in HTK were optimized; finally, the real time robust recognition of single notes, the HMM modeling of multi-note and the recognition of multi-note HMM model were successfully realized.
Keywords: Multi-note, endpoint detection, diverse dictionaries, HMM
modeling
