This project provides the spectral tilt modification of plain speech (speech produced in a quiet environment) to resemble corresponding Lombard speech (speech produced in a noisy environment) by Fractional Roll-off Filter. The spectral modification is employed in the three scales: whole utterance, each phone, and each frame.
Our research was presented in IEEE ASRU 2023 and was published as
Y. Sato and J. Villegas, "Spectral Tilt May Have a Smaller Impact on the Intelligibility of Speech in Noise," 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), Taipei, Taiwan, 2023, pp. 1-5, doi: 10.1109/ASRU57964.2023.10389688.
- MATLAB 9.13.0.2166757 (R2022b)
- macOS Ventura Version 13.31 (a)
This corpus contains 2,700 pairs of Lombard and plain utterances produced by 54 speakers. Each speaker utters sentences of the Grid corpus.