A Study on Restoration of Bone-Conducted Speech with LPC-Based Model (国際ワークショップ Frontiers in Speech and Hearing Research)

概要

論文の詳細を見る
In a noisy environment, bone-conducted speech seems to be more advantageous than normal noisy speech (i. e., air-conducted speech) because of its stability against surrounding noise. However the sound quality level of bone-conducted speech is very low, so restoring bone-conducted speech is challenging new topic in speech signal processing field. Some current models for restoring bone-conducted speech using cross-spectrum, long-term Fourier Transform till brought out the restored signal with the artifacts such as musical noise and echo. In previous study, we proposed an MTF-based model to solve these drawbacks and yield a restored signal with higher intelligibility. In this paper, we proposed another method, which based on the linear predictive coding (LPC) model. The LPC-based model is expected to yield restored signal that is not only better intelligibility for human hearing systems and also better for automatic speech recognition (ASR) systems. To evaluate the ability of this model to improve the voice quality, we compare it with previous models using one subjective measure and several objective measures. MOS (mean opinion score) and LSD (log-spectrum distortion) were selected as measures for estimating the improvement in intelligibility, which is useful for human hearing systems. Distances based on LPC coefficients and MFCCs (mel frequency cepstral coefficients) were selected for ASR systems. The measures prove that, LPC-based model is the better model for both human hearing and ASR systems. However the objective measures show out, the advantage of LPC-based model for ASR systems till depended on measured positions.
2006-03-20

A Study on Restoration of Bone-Conducted Speech with LPC-Based Model (国際ワークショップ Frontiers in Speech and Hearing Research)

スポンサーリンク

概要

著者

関連論文

スポンサーリンク