Publication Details
Title: Robust Speech Recognition Using Articulatory Information
Author: K. Kirchhoff
Group: ICSI Technical Reports
Date: August 1998
PDF: ftp://ftp.icsi.berkeley.edu/pub/techreports/1998/tr-98-037.pdf
Overview:
This report describes experiments in speech recognition using articulatory information. Previously, articulatory-based speech recognizers have exclusively been developed for clean speech; the potential of an articulatory representation of the speech signal for noisy test conditions, by contrast, has not been explored. Moreover, there have barely been attempts at systematically combining articulatory information with standard acoustic recognizers. This paper investigates these aspects and reports speech recognition results on a variety of acoustic test conditions for individual acoustic and articulatory speech recognizers, as well as for a combined system. On a continuous numbers recognition task, the acoustic system generally performs equal to, or slightly better than, the articulatory system, whereas the articulatory system shows a statistically significant improvement on noisy speech with a low signal-to-noise ratio. The combined system nearly always performs significantly better than either of the individual systems.
Bibliographic Information:
ICSI Technical Report TR-98-037
Bibliographic Reference:
K. Kirchhoff. Robust Speech Recognition Using Articulatory Information. ICSI Technical Report TR-98-037, August 1998
Author: K. Kirchhoff
Group: ICSI Technical Reports
Date: August 1998
PDF: ftp://ftp.icsi.berkeley.edu/pub/techreports/1998/tr-98-037.pdf
Overview:
This report describes experiments in speech recognition using articulatory information. Previously, articulatory-based speech recognizers have exclusively been developed for clean speech; the potential of an articulatory representation of the speech signal for noisy test conditions, by contrast, has not been explored. Moreover, there have barely been attempts at systematically combining articulatory information with standard acoustic recognizers. This paper investigates these aspects and reports speech recognition results on a variety of acoustic test conditions for individual acoustic and articulatory speech recognizers, as well as for a combined system. On a continuous numbers recognition task, the acoustic system generally performs equal to, or slightly better than, the articulatory system, whereas the articulatory system shows a statistically significant improvement on noisy speech with a low signal-to-noise ratio. The combined system nearly always performs significantly better than either of the individual systems.
Bibliographic Information:
ICSI Technical Report TR-98-037
Bibliographic Reference:
K. Kirchhoff. Robust Speech Recognition Using Articulatory Information. ICSI Technical Report TR-98-037, August 1998