README.md

metadata

license: apache-2.0
size_categories:
  - 100M<n<1B

NX Noise

（1）采样率 8000，int16
（2）所有 noise 噪音音频都是大于2秒的。
（3）speech 有少量是大于 2 秒，后续改成了最少 4 秒。
（4）noise 噪音段通过分类模型从音频中提取，可能会包含有语音。
（5）speech 是通过分类模型提取，不是纯净的语音。

speech 数据，可以根据 call_id 是否相同来判断是否是同一个人的讲话。可用于声纹识别的训练。
但是这些数据的时长比较短。

noise duration

duration;

language	count	duration (m)	duration (h)
ms-MY	31452	1754.67	29.24
en-PH	19502	1185.67	19.76
id-ID	81145	4913.79	81.9
pt-BR	123156	7366.67	122.78
TOTAL	255255	913248.21	253.68

two second duration;

language	count	duration (m)	duration (h)
ms-MY	31452	1398.17	23.3
en-PH	19502	953.03	15.88
id-ID	81145	3802.1	63.37
pt-BR	123156	5803.9	96.73
TOTAL	255255	717432.0	199.29

three second duration;

language	count	duration (m)	duration (h)
ms-MY	31452	819.7	13.66
en-PH	19502	615.0	10.25
id-ID	81145	2411.95	40.2
pt-BR	123156	3683.45	61.39
TOTAL	255255	451806.0	125.5

speech duration

duration;

language	count	duration (m)	duration (h)
en-PH	6673	673.81	11.23
id-ID	15623	1554.33	25.91
ms-MY	35771	3672.51	61.21
pt-BR	49306	4639.59	77.33
TOTAL	107373	632414.88	175.67

two second duration;

language	count	duration (m)	duration (h)
en-PH	6673	593.07	9.88
id-ID	15623	1328.5	22.14
ms-MY	35771	3219.33	53.66
pt-BR	49306	4004.63	66.74
TOTAL	107373	548732.0	152.43

three second duration;

language	count	duration (m)	duration (h)
en-PH	6673	508.3	8.47
id-ID	15623	1129.75	18.83
ms-MY	35771	2774.8	46.25
pt-BR	49306	3364.4	56.07
TOTAL	107373	466635.0	129.62