Closed Souvic closed 1 year ago
I am getting these scores on the test set if trimming is not done after 500 epochs.
--- Set: test --- CER: | s2s_st: 0.1787 | s2s_ut: 0.1920 | u2u_st: 0.1717 | u2u_ut: 0.1766 WER: | s2s_st: 0.3156 | s2s_ut: 0.3644 | u2u_st: 0.3217 | u2u_ut: 0.2973 ASV ACC: | s2s_st: 0.9500 | s2s_ut: 0.9600 | u2u_st: 0.9150 | u2u_ut: 0.9550 ASV COS: | s2s_st: 0.7993 | s2s_ut: 0.7960 | u2u_st: 0.7859 | u2u_ut: 0.7878
Is this ok, or should I rerun it several times to get to your scores because of the inherent uncertainty?
How do I calculate the average scores here? Specifically, what are s2sst and s2sut? s2s is already 'seen to seen' right?
okay, seen text and unseen text I guess..
That's right..!
I am getting these scores on the test set if trimming is not done after 500 epochs.
--- Set: test --- CER: | s2s_st: 0.1787 | s2s_ut: 0.1920 | u2u_st: 0.1717 | u2u_ut: 0.1766 WER: | s2s_st: 0.3156 | s2s_ut: 0.3644 | u2u_st: 0.3217 | u2u_ut: 0.2973 ASV ACC: | s2s_st: 0.9500 | s2s_ut: 0.9600 | u2u_st: 0.9150 | u2u_ut: 0.9550 ASV COS: | s2s_st: 0.7993 | s2s_ut: 0.7960 | u2u_st: 0.7859 | u2u_ut: 0.7878
Is this ok, or should I rerun it several times to get to your scores because of the inherent uncertainty?
How do I calculate the average scores here? Specifically, what are s2sst and s2sut? s2s is already 'seen to seen' right?