15th Annual Conference of the International Speech Communication Association

September 14-18, 2014

Spectral Tilt Modelling with GMMs for Intelligibility Enhancement of Narrowband Telephone Speech

Emma Jokinen, Ulpu Remes, Marko Takanen, Kalle Palomäki, Mikko Kurimo, Paavo Alku

Aalto University, Finland

In mobile communications, post-processing methods are used to improve the intelligibility of speech in adverse background noise conditions. In this study, post-processing based on modelling the Lombard effect is investigated. The study focuses on comparing different spectral envelope estimation methods together with Gaussian mixture modelling in order to change the spectral tilt of speech in a post-processing algorithm. Six spectral envelope estimation methods are compared using objective distortion measures as well as subjective word-error rate and quality tests in different near-end noise conditions. Results show that one of the envelope estimation methods, stabilised weighted linear prediction, yielded statistically significant improvement in intelligibility over unprocessed speech.

Full Paper

Bibliographic reference.  Jokinen, Emma / Remes, Ulpu / Takanen, Marko / Palomäki, Kalle / Kurimo, Mikko / Alku, Paavo (2014): "Spectral tilt modelling with GMMs for intelligibility enhancement of narrowband telephone speech", In INTERSPEECH-2014, 2036-2040.