Exemplar-based speech enhancement and its application to noise-robust aut omatic speech recognition - Demonstrations

The following audio examples demonstrate the signals produced by the algorithm proposed in this paper presented in the CHiME workshop. The original signals are from the CHiME dataset. The enhanced signals produced by both the adaptive and fixed noise dictionaries are available below. The "SNR" in the table below refers to the signal-to-noise ratios of the original signal, as defined in the original dataset.

Original signal SNR Signal enhanced with an adaptive noise dictionary Signal enhanced with a fixed noise dictionary
orig_m6dB_s13_lgid8n.wav -6 dB adaptive_m6dB_s13_lgid8n.wav fixed_m6dB_s13_lgid8n.wav
orig_m6dB_s15_pban4n.wav -6 dB adaptive_m6dB_s15_pban4n.wav fixed_m6dB_s15_pban4n.wav
orig_m6dB_s8_sbbg6s.wav -6 dB adaptive_m6dB_s8_sbbg6s.wav fixed_m6dB_s8_sbbg6s.wav
orig_0dB_s15_brbr6n.wav 0 dB adaptive_0dB_s15_brbr6n.wav fixed_0dB_s15_brbr6n.wav
orig_0dB_s3_pgij8p.wav 0 dB adaptive_0dB_s3_pgij8p.wav fixed_0dB_s3_pgij8p.wav
orig_0dB_s8_sbis8a.wav 0 dB adaptive_0dB_s8_sbis8a.wav fixed_0dB_s8_sbis8a.wav
orig_6dB_s23_bbbx6n.wav 6 dB adaptive_6dB_s23_bbbx6n.wav fixed_6dB_s23_bbbx6n.wav
orig_6dB_s5_lrikzp.wav 6 dB adaptive_6dB_s5_lrikzp.wav fixed_6dB_s5_lrikzp.wav
orig_6dB_s7_swau4p.wav 6 dB adaptive_6dB_s7_swau4p.wav fixed_6dB_s7_swau4p.wav

Demonstrations main page

- Tuomas Virtanen, tuomas.virtanen@tut.fi