LinJinghuidev

自己的数据，训练效果不理想

2

拿aishell3的数据集训练，loss下降的很快，模型run2000轮就能输出较为清晰的语音。用自己收集来的语音去训练，收敛很慢且输出结果不太理想。 自己的数据频谱清晰无杂音，不是很明白为什么效果和aishell差这么多，请指教

about offsets

1

What's the use of offsets and how to get the value [ [-42.198200,91.614723,-40.067841], [ 0.103456,1.857829,10.548506], [43.499992,-0.000038,-0.000002], [42.372192,0.000015,-0.000007], [ 17.299999,-0.000002,0.000003], [0.000000,0.000000,0.000000], [0.103457,1.857829,-10.548503], [43.500042,-0.000027,0.000008], [42.372257,-0.000008,0.000014], [17.299992,-0.000005,0.000004], [0.000000,0.000000,0.000000], [6.901968,-2.603733,-0.000001], [12.588099,0.000002,0.000000], [12.343206,0.000000,-0.000001], [25.832886,-0.000004,0.000003], [11.766620,0.000005,-0.000001],...

about R=rot, t=0

3

about R=rot, t=0 (rot = np.array([0.14070565, -0.15007018, -0.7552408, 0.62232804], dtype=np.float32)) R and T are parameters that determine the camera Does this mean that when I create a dataset, the position...