Contact : Sara Grassi

Speech signal compression, aiming to encode and decode the signal, finds many applications such as fixed or mobile digital telephony, packet network transmission (Internet), videoconferencing, radio and television, and various storage applications.

Narrowband speech is defined as speech filtered in continuous time in the frequency band [300 - 3400 Hz], and then sampled at 8 kHz. Narrowband speech is usually called "speech" in the telephony band and is used by the second generation of mobile telephony (2G, in Europe GSM phase 2). Narrowband speech coding makes it possible to obtain a reconstructed speech signal with sufficient quality for telephony. However, the reconstructed signal does not have a natural consonance, and may sound metallic or even synthetic. It is sometimes difficult to distinguish two fricative consonants such as the "S" or "F" in such a signal.

Wideband speech is defined as speech filtered in continuous time, in the frequency band [50 7000 Hz], and then sampled at 16 kHz. The interest of processing a wideband speech signal is to obtain a reconstructed signal that is clearer, more natural and intelligible than with narrowband speech.

There is a great diversity of speech coding techniques. Waveform coders, use for example Pulse Code Modulation (PCM) or Adaptive Differential PCM (ADPCM). These coders make it possible to obtain an excellent quality of the reconstructed signal but generally result in a low compression ratio. Other techniques, such as those based on CELP (Code Excited Linear Prediction), use a coding scheme carrying out an analysis-by-synthesis and are based on speech production and perception models. Such coders result in a reconstructed signal of good quality with a higher compression ratio.

ESPLAB has worked for a long time in software and hardware implementation of various narrowband speech coding functionalities. With the diffusion of multimedia technologies and the introduction of services for the 3rd and 4th generations of mobile telephony, wideband speech coding became a prevailing research field and ESPLAB has developed a proprietary multi-rate wideband speech coder.

Example of current development:
Wideband speech coding for storing the compressed speech in a portable device in order to realize a remote decompression on a PC. The decompressed speech is then used for an automatic transcription into text.

Publications :

• Giuseppina Biundo-Lotito, Codage à débit variable de la parole en bande élargie, PhD Thesis, University of Neuchâtel, IMT, June 2003.
• G. Biundo, M. Ansorge, F. Pellandini et P.-A. Farine, "Perceptual weighting for ACELP wideband speech coder", dans Proc. of 4th COST 276 Workshop on information and knowledge management for integrated media communication, pp. 105-110, Bordeaux, France, Mars-Avril 2003.
• G. Biundo, S. Grassi, M. Ansorge, F. Pellandini et P.-A. Farine, "Design techniques for spectral quantization in wideband speech coding", dans Proc. of 3rd COST 276 Workshop on information and knowledge management for integrated media communication (CD-ROM), Budapest, Hongrie, Oct. 2002.
• G. Biundo, S. Grassi, M. Ansorge et F. Pellandini, "Spectral quantization for wideband speech coding", dans Proc. of 1st COST 276 Workshop on information and knowledge management for integrated media communication (CD-ROM), Leganés (Madrid), Espagne, Nov. 2001.
• S. Grassi, M. Matthey, G. Biundo, A. Dufaux, M. Ansorge, F. Pellandini, "A Low-Complexity Variable Bit Rate Speech Coder for Portable Storage Applications", Proc. of Signal Processing 2000 Workshop, Poznan, Poland, Oct. 13th, 2000, pp. 7-12.
• S. Grassi, M. Ansorge, F. Pellandini, "LSP Calculation Methods for Applications to Speech Coding", Proc. of the Signal Processing 99 Workshop, Poznan, Poland, October 8, 1999, pp. 61-66.
• S. Grassi, M. Ansorge, F. Pellandini, "An Algorithm for Fast Direct Calculation of Quantized LSP Parameters", Proc. of the Fifth Bayona Workshop on Emerging Technologies in Telecommunications, COST 254, Bayonna, Spain, September 6-8, 1999, pp. 158-162.
• S. Grassi, Optimized implementation of speech processing algorithms, Thèse éditée par la Faculté des Sciences de l'Université de Neuchâtel, Neuchâtel, 1998.