EURASIP Journal on Audio, Speech, and Music Processing

Table 2 Test the effect of noise on model performance

From: Multi-task deep cross-attention networks for far-field speaker verification and keyword spotting

	SNR	Seen			Unseen
	SNR	EER (%)	minDCF	Acc (%)	EER (%)	minDCF	Acc (%)
Noise	− 5 dB	4.11	0.193	90.72	8.07	0.406	87.85
	0 dB	1.34	0.947	94.70	4.92	0.266	91.95
	5 dB	0.63	0.037	96.49	3.68	0.206	93.75
	10 dB	0.27	0.019	97.45	2.81	0.175	94.65
	15 dB	0.20	0.014	97.85	2.44	0.158	95.17
Music	− 5 dB	5.13	0.251	89.83	9.54	0.453	86.97
	0 dB	1.49	0.082	94.68	5.14	0.266	91.75
	5 dB	0.53	0.032	96.68	3.55	0.201	94.03
	10 dB	0.26	0.014	97.76	2.53	0.162	94.87
	15 dB	0.17	0.007	98.04	2.24	0.143	95.35

Back to article page