In this paper we propose a new musical instrument recognition technique based on hidden markov model (HMhf). Spectral envelop is the key information of instrument characteristic and timbre. We decompose an instrument sound into sinusoidal components (harmonics) and noise component and estimate the amplitudes of harmonics component. We want to express spectral envelop effectively using estimated amplitudes, therefore we define three kinds of features and apply recognition procedure to each feature. The HAfM model used in this paper is continuous single gaussian output HAfM. To evaluate the performance of recognition technique, we apply proposed technique to classify real instrumental sound of J4UMS (AfacGill University Master Samples). The recognition success ratio is more than 70% .
[1]
J C Brown,et al.
Feature dependence in the automatic identification of musical woodwind instruments.
,
2001,
The Journal of the Acoustical Society of America.
[2]
Xavier Rodet,et al.
Timbre Recognition with Combined Stationary and Temporal Features
,
1998,
ICMC.
[3]
Youngmoo E. Kim,et al.
Musical instrument identification: A pattern‐recognition approach
,
1998
.
[4]
Biing-Hwang Juang,et al.
Fundamentals of speech recognition
,
1993,
Prentice Hall signal processing series.
[5]
S. Kay.
Fundamentals of statistical signal processing: estimation theory
,
1993
.