Is it planned to support Conformer-CTC models on the Jarvis toolkit?

AI & Data Science Deep Learning (Training & Inference) Riva
1

I’ve tried using Conformer model in Jarvis v1.3, but I get the following error when loading the model:
Снимок

2

Hi @sltl
“Network must have at least one output”
It seems to be model issue. Could you please check if generated onnx model is valid using below command:

import sys
import onnx
filename = yourONNXmodel
model = onnx.load(filename)
onnx.checker.check_model(model)

Also, could you please share the model, config and script that you are using so we can help better?

Thanks

3

I used the standard nemo model “stt_en_conformer_ctc_large” STT En Conformer-CTC Large | NVIDIA NGC
This model was converted from .nemo to .ejrvs format using nemo2jarvis-1.3.0b0-py3-none-any.whl script
I tried to run Jarvis build with parameters like for Citrinet model, build finished with no errors:

jarvis-build speech_recognition |
/servicemaker-dev/stt_en_conformer_ctc_large.jmir |
/servicemaker-dev/stt_en_conformer_ctc_large.ejrvs |
–chunk_size=0.8 |
–padding_size=1.6 |
–ms_per_timestep=80 |
–lm_decoder_cpu.asr_model_delay=-1 |
–featurizer.use_utterance_norm_params=False |
–featurizer.precalc_norm_time_steps=0 |
–featurizer.precalc_norm_params=False |
–lm_decoder_cpu.decoder_type=greedy |

The Onnx model was created and the check for “onnx.checker.check_model ()” was successful.

4

Are Nvidia Nemo Conformer-CTC models supported by the current version of Jarvis?

5

Hi @sltl
Currently it’s not supported, may be in future support will be added.
Please stay tune for updates.

Thanks