| 摘要 |
PURPOSE:To provide a method by which a transmitting error of a code hardly occurs and a residual signal is efficiently quantized when encoding is performed by a tranformation encoding method of an MDCT means. CONSTITUTION:An acoustic input signal is linearly prediction analyzed by a means 37, the result is quantized, a frequency characteristic outline of an input signal is calculated from the quantized prediction coefficient by a means 41. The input signal is transformed with transformation discrete cosine by a MDCT means 36 and is made a frequency region signal, the signal is normalized in a frequency outline by a means 42, and made a flat residual signal. Power of the residual signal is normalized, vector quantization weighed with the frequency outline is performed by a means 46. In this quntization, both inputs are distributed successively to plural small groups by one sample, and wlighted vector quantization is performed for each small group. |
