METHOD OF DISCRIMINATING BACKGROUND SOUND/VOICE, METHOD OF DISCRIMINATING VOICE SOUND/UNVOICED SOUND, METHOD OF DECODING BACKGROUND SOUND

摘要

PROBLEM TO BE SOLVED: To offer a method permitting to precisely discriminate background sound/voice even under a situation in which background noise has a large power or under background noise in which spectrum of running sound of a car, train, voices of others, etc., is not flat. SOLUTION: From the information of a calculated frame power obtained by supplying a digital input signal inputted from an input terminal 101 to a featured value calculation part 102, an estimated frame power obtained by a calculated LSP coefficient and an estimated featured value updating part 104, and an estimated LSP coefficient, a background sound/voice judgment part 103 judges whether an input signal is a background sound or a voice, and then, the estimated featured value updating part 104 updates the estimated frame power and estimated LSP coefficient by using the frame power and LSP coefficient obtained by the featured value calculation part 102 for preparing for the next frame.