Estudo sobre transientes em sinais de fala e mùsica

Abstract

Nowadays, computer systems require the modeling of musical and speech signals for several purposes: synthesis, computational hearing, acoustic analysis, systematic musicology, transformation, composition, and many others. The quality modeling synthesis of music and speech signals is a complex task, which is still presented as a challenge.This is particularly true when low dimensional parametric representation, low computational cost, intelligibility and naturalness are aimed.Speech and musical signals have transient states of vibration, such as musical instrument note attacks and speech bursts in stop consonants. Due to its impulsive characteristics, transients are like snapshots of the vocal tract and musical instruments, being important for the perception of timbre and recognition of the sound source. Many of the modeling techniques of musical and speech signals are not efficient attransient parts. The separation of transients from deterministic and stochastic signal components represent a significant improvement in modeling synthesis flexibility. This study is devoted to the analysis, modeling, and auditory perception measurements of speech and music transients. In the research carried out, Transient Modeling Synthesis (TMS) is used to model the transient components of speech and musical signals. Next, TMS is evaluated and compared to traditional sinusoidal based Spectral Modeling Synthesis (SMS). The results of a phoneme recognition experiment and a quality MOS (Mean Opinion Score)test are used to measure the importance of an adequate modeling of transients as bursts in stop consonants. Compared to the intelligibility of 98% obtained for the original utterances, TMS modeling attained 95%, which are significantly higher than the 87% attained withSMS modeling. It was also observed that removing the transient component reduces the intelligibility to 79%. Finally, by enabling the separation of transient components, TMS allows the definitionof an index to measure the ratio between the energy of original and of transient signal components. This index, called Index of Transience, has been evaluated. The values obtained were then compared in a test and applied to different musical instrument notes.