Speech Enhancement Using a Self-Attentive Multi-Stage Temporal Convolutional Network

Ju Lin, Adriaan J. van Wijngaarden, Kuang-Ching Wang and Melissa C. Smith

Examples from VCTK

Id	Noisy	1stage SA-TCN	2stage SA-TCN	3stage SA-TCN	4stage SA-TCN	5stage SA-TCN	Clean
p232_005
p232_010
p232_013
p232_014
p232_032

Examples from Librispeech

Babble noise, SNR = -5 dB
Id	Noisy	CRN	Complex-CNN	DARCN	5stage-SA-TCN	Clean
sample1
sample2

Babble noise, SNR = 0 dB
Id	Noisy	CRN	Complex-CNN	DARCN	5stage-SA-TCN	Clean
sample1
sample2

Babble noise, SNR = 5 dB
Id	Noisy	CRN	Complex-CNN	DARCN	5stage-SA-TCN	Clean
sample1
sample2

Kitchen noise, SNR = -5 dB
Id	Noisy	CRN	Complex-CNN	DARCN	5stage-SA-TCN	Clean
sample1
sample2

Kitchen noise, SNR = 0 dB
Id	Noisy	CRN	Complex-CNN	DARCN	5stage-SA-TCN	Clean
sample1
sample2

Kitchen noise, SNR = 5 dB
Id	Noisy	CRN	Complex-CNN	DARCN	5stage-SA-TCN	Clean
sample1
sample2