Cognitive Superiority of Phase Error to Amplitude Envelope in Sound

1オクターブに帯域制限した音響波形の極大値と極小値の情報があれば、元の波形を組み立てて情報を再現することができる。位相情報と振幅包絡はそれぞれ音源定位と音声情報の運び手の1つと関係付けられているので、計測した極値の位相と振幅方向の誤差のどちらが合成音の音質に重大な影響を及ぼすかについては関心が持たれるところである。そこで位相誤差と振幅誤差を独立に与えた合成音の主観評価をしたところ、2種類の合成音の波形と元の波形との二乗誤差は等しいにもかかわらず、位相誤差が有意に音質の劣化を招いていることを観察した。加えて極値を最小二乗推定することで位相誤差を4%未満に抑制すれば、合成音の音質が改善された。以上より、聴性認識は振幅より位相の検出に敏感にできており、位相誤差を算出すれば合成音の音質を見積もるための指標に利用できることが示唆された。It has been reported that the maximum and minimum data of the narrow-band (one octave) acoustic waveforms played a pivotal role in the sound reproduction by using the sinusoidal interpolation between two successive extrema. Although two indices of phase and quantization errors of the extrema have closely related to the perception of sound localization, and the form of amplitude envelope, respectively, which has regarded as a principal carrier of speech waveforms, the interaction between the two has remained unclear as to which is a critical factor in the auditory cognitive system. We produced two types of synthesized sounds, manipulating either time or amplitude of the extrema under condition that two sum squared errors between an unmodified sound and the two syntheses were mutually close. Scheffe's pair test has reported that the quality of synthesized 3-s speech and music with phase error was significantly lower than that of the corresponding sounds with quantization error, suggesting the cognitive superiority of phase error to noise in amplitude envelope. Besides, synthesized sounds with estimated extrema by using a least-square fit maintained almost the same quality of unprocessed sounds, when phase error was within the margin of four to nine percent, supporting the notion of an index, sensitive to the sound quality