Tomasz Zernicki: Sinusoidal modeling of music in the context of audio bandwidth extension and signal compression.
Sinusoidal modeling is well established signal processing frameworks applicable to speech and audio analysis, time and pitch scaling, enhancement, restoration, source separation, automatic recognition, watermarking, compression, and synthesis. Zernicki will demonstrate his work on hybrid sinusoidal modeling Matlab toolbox as well as its usage in the context of audio compression.
This work is related to low-bitrate audio coding (16 kb/s) using bandwidth extension techniques. Two different bandwidth extension techniques will be demonstrated. First technique was implemented as an extension to MPEG-4 AAC HE. This technique exploits jointly spectral band replication and sinusoidal modeling. Second technique uses frequency scaling of narrowband signals for reconstruction of high frequency tonal components. Compression efficiency has been assessed in extensive experimental tests with usage of standard audio excerpts. This quality assessment has been performed using standard listening test according to BS.1534 ITU-R Recommendation.
Results obtained during listening tests proved that both prepared techniques outperform standard MPEG-4 AAC HE technique. This improvement is significant for excerpts with strong tonal components. Moreover, proposed techniques has been integrated with new audio compression standard MPEG-D USAC.
Tomasz Żernicki (http://www.linkedin.com/in/tomaszzernicki) is a co-founder and a Senior Research Scientist at NextDayLab (www.nextdaylab.com) and Zylia R&D. companies are focus on applied research in the field on multimedia and networking. His professional interests concentrate on: audio and video compression algorithms, spatial sound processing, audio bandwidth extension, parametric sound representation and modeling.
He received PhD degree (2010) from Poznan University of Technology (Poland) in the field of Electronics and Telecommunications. In the past he worked as a Senior Research Scientist at Telcordia Poland Sp. z o.o. Applied Research Center and as an assistant at Poznan University of Technology. He took part in several industry-oriented research projects related to digital television as well as audio and video MPEG-4 codec implementation for embedded platforms.
As an audio expert he takes an active role in the Moving Picture Expert Group (MPEG) standardization committee work. He proposed an extension for audio compression standard MPEG-D Unified Speech and Audio Coding (USAC). This technique exploits jointly Spectral Band Replication (SBR) and sinusoidal modeling in order to improve encoding of high frequency tonal components for low bit rates (16 – 24 kb/s).