Hi, I’m using the version of DeepSpeech from (https://github.com/mozilla/DeepSpeech). I hope to show the WER in training process, so I tried to add an op in function " calculate_mean_edit_distance_and_loss", like:
"
decoded, _ = tf.nn.ctc_beam_search_decoder(logits, batch_seq_len, FLAGS.beam_width)
wer = tf.edit_distance(tf.cast(decoded[0], tf.int32), batch_y, normalize=True)
avg_wer = tf.reduce_mean(wer)
" . However, it will make the training process too slow (five hour an epoch). I notice a parameter “–display_step” used by others, but I don’t find it in the version of DeepSpeech source code download from https://github.com/mozilla/DeepSpeech. Is there anyone could give some suggestions about this? Thanks a lot !