I used the OpenSeq2Seq container (nvcr.io/nvidia/tensorflow:19.05-py3) to train a DeepSpeech model using my own data. I tried to obtain the timestamps following the steps from here 1. Due the fact that the config _‘infer_logits_topickle’: True in the decoder params was not recognized in that container version, I switched to the latest one (nvcr.io/nvidia/tensorflow:19.10-py3). But when I tried to run ./scripts/dump_to_time.py I got the following error:
Warning: no calibration parameters were provided, using zeros instead
Traceback (most recent call last):
File "./scripts/dump_to_time.py", line 51, in
letters, starts, ends = ctc_greedy_decoder(results[r], vocab, step_size, 28, start_shift, end_shift)
File "/workspace/nvidia-examples/OpenSeq2Seq/open_seq2seq/utils/ctc_decoder.py", line 30, in ctc_greedy_decoder
output += wordmap[idx]
KeyError: 33
Please guide me to a solution to obtain the timestamps in OpenSeq2Seq using a DeepSpeech model. Could be an issue because I trained with a different container version than the latest one, which I’m using for inference?
Hello,
I used the OpenSeq2Seq container (nvcr.io/nvidia/tensorflow:19.05-py3) to train a DeepSpeech model using my own data. I tried to obtain the timestamps following the steps from here 1. Due the fact that the config _‘infer_logits_topickle’: True in the decoder params was not recognized in that container version, I switched to the latest one (nvcr.io/nvidia/tensorflow:19.10-py3). But when I tried to run ./scripts/dump_to_time.py I got the following error:
Warning: no calibration parameters were provided, using zeros instead Traceback (most recent call last): File "./scripts/dump_to_time.py", line 51, in
letters, starts, ends = ctc_greedy_decoder(results[r], vocab, step_size, 28, start_shift, end_shift)
File "/workspace/nvidia-examples/OpenSeq2Seq/open_seq2seq/utils/ctc_decoder.py", line 30, in ctc_greedy_decoder
output += wordmap[idx]
KeyError: 33
Please guide me to a solution to obtain the timestamps in OpenSeq2Seq using a DeepSpeech model. Could be an issue because I trained with a different container version than the latest one, which I’m using for inference?
Thank you!