3 ms·

You 100% tuned your hyper parameters and LM for librespeech, and only using librispeech data probably helped the result. When you start training systems for LVC

You 100% tuned your hyper parameters and LM for librespeech, and only using librispeech data probably helped the result. When you start training systems for LVCSR it's normal for the WER to go up on some test sets. (This is why you can never expect a commercial system to get SOTA on librispeech, despite it being much better system in reality)

gok 7y ago
But these results are about in line with other CTC systems. The original Baidu paper [1] had 7.89% with original Deep Speech and 5.33% with "Deep Speech 2" where they threw more complicated networks at it and added another 11K hours of external speech traingin data. [1] https://arxiv.org/pdf/1512.02595v1.pdf https://arxiv.org/pdf/1512.02595v1.pdf