Spectral Tilt Estimation for Speech Intelligibility Enhancement Using RNN Based on All-Pole Model
Speech intelligibility enhancement is extremely meaningful for successful speech communication in noisy environments. Several methods based on Lombard effect are used to increase intelligibility. In those methods, spectral tilt has been suggested to be a significant characteristic to produce Lombard speech that is more intelligible than normal speech. All-pole model computed by some methods has been used to capture the accurate spectral tilt of high-quality speech, but they are not appropriate for the spectral tilt estimation of telephone speech. In this paper, recurrent neural networks (RNNs) are used to estimate the tilt of telephone speech in German and English. RNN-based spectral tilt estimation show the robustness on the change of the all-pole model order and phonation type for narrow and wideband speech. Mean squared error (MSE) of spectral tilt estimation using RNN-based method is increased by about 26.20% in narrow speech and 19.49% in wideband speech comparing to the DNN-based measure.
KeywordsSpectral tilt All-pole model RNN
This work is supported by National Key Program of China (No. 2017YFB1002803) and National Nature Science Foundation of China (No. U1736206, No. 61801334, No. 61762005).
- 2.Sauert, B., Vary, P.: Near end listening enhancement optimized with respect to speech intelligibility index. In: 17th European Signal Processing Conference, pp. 1844–1848. IEEE (2009)Google Scholar
- 3.Schepker, H., Rennies, J., Doclo, S.: Improving speech intelligibility in noise by SII-dependent preprocessing using frequency-dependent amplification and dynamic range compression. In: Proceedings of Interspeech, pp. 3577–3581. ISCA, Lyon (2013)Google Scholar
- 5.Petko, P.N., Stylinaou, Y.: Adaptive gain control and time warp for enhanced speech intelligibility under reverberation. In: IEEE International Conference on Acoustic, Speech and Signal Processing (IASSP), New Orleans, pp. 691–695. IEEE (2017)Google Scholar
- 6.Zorilâ, T.C., Kandia, V., Stylianou, Y.: Speech-in-noise intelligibility improvement based on spectral shaping and dynamic range compression. In: Proceedings Interspeech, pp. 635–638. ISCA, Portland (2012)Google Scholar
- 8.Jokinen, E., Remes, U., Takanen, M., et al: Spectral tilt modelling with GMMs for intelligibility enhancement of narrowband telephone speech. In: Proceedings of Interspeech, pp. 2036–2040. ISCA, Singapore (2014)Google Scholar
- 9.Jokinen, E., Remes, U., Alku, P.: The use of read versus conversational Lombard speech in spectral tilt modeling for intelligibility enhancement in near-end noise conditions. In: Proceedings of Interspeech, pp. 2771–2775. ISCA, San Francisco (2016)Google Scholar
- 12.Bronkhorst, A.W.: The cocktail party phenomenon: a review of research on speech intelligibility in multiple-talker conditions. Acta Acust. United Acust. 86(1), 117–128 (2000)Google Scholar
- 20.Airaksinen, M., Story, B., Alku, P.: Quasi closed phase analysis for glottal inverse filtering. In: 14th Annual Conference of the International Speech Communication Association, pp. 143–147. ISCA, Lyon (2013)Google Scholar
- 23.Sofoklis, K., Okko, R., Pavvo, A.: Evaluation of spectral tilt measures for sentence prominence under different noise conditions. In: Proceedings of Interspeech, pp. 3211–3215. ISCA, Stockholm (2017)Google Scholar
- 24.Lopez, A.R., Seshadri, S., Juvela, L., et al.: Speaking style conversion from normal to Lombard speech using a glottal vocoder and Bayesian GMMs. In: Proceedings of Interspeech, pp. 1363–1367. ISCA, Stockholm (2017)Google Scholar
- 26.Sołoducha, M., Raake, A., Kettler, F., Voigt, P.: Lombard speech database for German language. In: Proceedings of DAGA, Aachen (2016)Google Scholar