Standardized Frontends
for Speech Recognition
Data Bases and Experiments
on Robust Recognition
Download
-
Set of scripts to perform the Aurora-2 experiment with the advanced ETSI front-end
-
FaNT – Filtering and Noise Adding Tool
To create the Aurora-2 and the Aurora-4 data, noise signals have been artificially added to the clean TIDigits (downsampled at 8 kHz) respectively to the WSJ (wall street journal) speech data (downsampled at 8 or 16 kHz). The software tool for noise adding and filtering is available as C program:
-
Aurora-5 - Meeting Recorder Digits
Recorded in hands-free mode at the International Computer Science Institute in Berkeley