Voice Adaptation from Pre-processed Voice Profile with Sound-indexed Library Search

摘要

This paper proposes a two-step real-time voice adaptation system in the field of speech processing. Step one combines recording and pre-processing to construct a voice profile using a helper program to ensure the data is trained within suggested parameters. Secondly is the real-time input of the voice's adaptation to a target voice. This data is put through our proposed methods of Abstract Normalization and then Sound-indexing, which maps each sound into a library for the given voice. The speaker's power is measured on input and then artificially applied to the output data to improve the sound quality. The second part is the actual voice adaptation where the input speaker's voice goes through a similar data conversion before compared against the voice profiles'. Ultimately, the method is tested using an English voice profile. Experimental results show that the matches are easily done with simple input, but complicated, longer input proves more difficult.