3 ms·
I haven't trained on LibriSpeech exclusively, but yes, the perf on LibriSpeech dev is quite bad, around ~60.0 WER. If the poor alignment of yt captions is the i
by iceychris 6y ago
I haven't trained on LibriSpeech exclusively, but yes, the perf on LibriSpeech dev is quite bad, around ~60.0 WER. If the poor alignment of yt captions is the issue, maybe concatenating multiple samples helps a bit.
- lunixbochs 6y agoYou should consider realignment; maybe start with something like DSAlign or my wav2train project.