Project

• Signal processing will involve non-linear speaker and channel adaptation by finding a common low dimensional mapping of training data, based on the J-RASTA signal processing approach. Knowledge of the Multi-band Recognition paradigm will be experimented with and exploited.
• Variability as a function of global and local speaking rate will be incorporated into the acoustic model. We will exploit discriminant HMM technology developed by Bourlard and Morgan, using transition probabilities that are dependent on acoustics (and in this case rate).

People

Jordan Cohen - IDA (leader)

Herve Bourlard - Faculte Polytechnique de Mons(BE) / ICSI

Hynek Hermansky - Oregon Graduate Institute

Nikki Mirghafori - ICSI

Nelson Morgan - ICSI

Mark Ordowski - NSA

Christophe Ris - Faculte Polytechnique de Mons(BE)

Sangita S. Tibrewala - Oregon Graduate Institute

Resources

Spert board and Quicknet (fast training of large neural networks).

The Speech Training and Recognition Unified Tool (see the STRUT homepage in Mons, Belgium)

Report

Description of the experiments

Results

Final Report (Word document)

---

Last modified on July 18, 1996

Christophe Ris <ris@cspjhu.ece.jhu.edu >