Hardware - GPU (T4)
Hardware - CPU
Operating System
Riva Version: 2.6.0
TLT Version (if relevant)
How to reproduce the issue ? (This is for errors. Please share the command and the detailed log here)
Hello everyone again!
After the riva model deployment, I’m generating transcriptions from my audio files. It’s working fine.
I wanted head to head comparison of the raw .nemo
model with the riva model repositories. But I’ve noticed that, in few cases like shorter audios(not applicable for all the shorter audios) I’m not getting transcriptions, where I was getting from raw .nemo
model. I’ve tried with both Quick Start Scripts way and Docker way, but with no luck. Both of these give same results, as expected.
Here’s the command I used to build rmir
model:
riva-build speech_recognition /servicemaker-dev/<output name of rmir model> /servicemaker-dev/<riva model name> --name=conformer-bn-BD-asr-streaming --featurizer.use_utterance_norm_params=False --featurizer.precalc_norm_time_steps=0 --featurizer.precalc_norm_params=False --ms_per_timestep=40 --endpointing.start_history=200 --nn.fp16_needs_obey_precision_pass --endpointing.residue_blanks_at_start=-2 --chunk_size=0.16 --left_padding_size=1.92 --right_padding_size=1.92 --decoder_type=flashlight --decoding_language_model_binary=<lm_binary> --decoding_vocab=<decoder_vocab_file> --flashlight_decoder.lm_weight=0.2 --flashlight_decoder.word_insertion_score=0.2 --flashlight_decoder.beam_threshold=20. --language_code=bn-BD
What could be the underlying cause for not receiving transcriptions after the model transformation? Is it related to VAD(voice activity detection) somehow? In that case, how to control this in riva model built?
Thanks in advance!