保持听觉一致性的音频缩放方法

An Audio Scaling Method Keeping Auditory Consistency

提出了一种以保持听觉一致性为目标的音频缩放方法。首先利用基音分析技术将整个音频波形分割为若干基音周期,然后以半音为单位将连续的基音周期组织成音频段落,并在各个段落内部按指定比例进行缩放。由于减少或增加的音频片段是以基音周期为标准的,并挑选与前后周期相似程度最大的基音周期进行缩放,从而较好地保持了原音频波形的基本要素(音强、音高、音色、波形包络等),使听觉失真达到最小。由于算法简单,能够满足实时处理的要求。实验表明,在±20%的缩放幅度内,基本没有听觉失真的现象发生。

An audio scaling method which produces minimal audible artifacts is proposed in this paper. The audio waveform is divided into several pieces according to the pitch period, and then the successive pieces with the same semitone are grouped into one segment. The scaling operations are performed on all these segments by the pre-defined scaling factor. All the deleted （or duplicated） pieces are based on pitch period, and are most similar to their neighbors, so the essential elements of the waveform （such as loudness, pitch, timbre, shape, etc. ） are largely saved. Due to its simplicity, the algorithm can be implemented in real-time environment. Experiments show that, in most cases, no audible artifacts are introduced within the moderate scope （up to 20%）.