INTERSPEECH 2004 - ICSLP
A multi-mode harmonic transform coding (MMHTC) for speech and music signals is proposed. Its structure is organized as a linear prediction model with an input of harmonic and transform-based excitation. The proposed coder also utilizes harmonic prediction and an improved quantizer of excitation signal. To efficiently quantize the excitation of music signals, the modulated lapped transform (MLT) is introduced. In other words, the coder combines both the time domain (linear prediction) and the frequency domain technique to achieve the best perceptual quality. The proposed coder showed better speech quality than that of the 8kbps QCELP coder at a bit-rate of 4kbps.
Bibliographic reference. Kim, Jong-Hark / Shin, Jae-Hyun / Lee, In-Sung (2004): "Multi-mode harmonic transfrom excitation LPC coding for speech and music", In INTERSPEECH-2004, 2669-2672.