Originally published at: https://developer.nvidia.com/blog/building-transcription-and-entity-recognition-apps-using-jarvis/
In the past several months, many of us have grown accustomed to seeing our doctors over a video call. It’s certainly convenient, but after the call ends, those important pieces of advice from your doctor start to slip away. What was that new medication I needed to take? Were there any side effects to watch…
Can sample rate for Jarvis ASR be 8000hz?
Thanks for your question! Yes, Jarvis supports 8 kHz input for ASR. The server will automatically upsample to 16 kHz. Just make sure you specify the sample rate as 8000 in the recognitionConfig being sent in the request.
@jwitsoe
Hello
We were able to run the rive docker client service
However how exactly to replicate the demo in the article https://developer.nvidia.com/blog/building-transcription-and-entity-recognition-apps-using-riva/#entry-content-comments ?
Thanks for your interest in the demo. Glad to hear you’ve been able to run the docker client. We have included instructions on running that demo from the docker here: Riva Contact — NVIDIA Riva
Cheers,
Chris
1 Like
@cparisien
Hi, thank you for your reply
Could you extend on how to start the callcenter with Medical NER German model, please?
AV
Hi @Andrey1984,
We don’t have a medical NER model in German, and unfortunately I’m not immediately aware of a good dataset you could use. A quick search turned up this one, which might be worth a try – I can’t comment on how well it might suit your application, and you’d likely have to do some data preprocessing.
If you do find a suitable dataset, you could use TAO to train or adapt an appropriate model:
https://docs.nvidia.com/tao/tao-toolkit/text/nlp/token_classification.html
Then once you have that model, you can deploy it in Riva:
https://docs.nvidia.com/deeplearning/riva/user-guide/docs/custom-model-deployment.html
I hope that helps!
Chris
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.
@cparisien
Hi, so what are the existing options for german ASR?
Is there more support by now for the Medical NER ?
How do we integrate GitHub - frankkramer-lab/GERNERMED: GERNERMED is the first open neural NER model for medical entities designed for German data. ?
Could you extend on steps to get it as a proof of concept implemented by using “TAO to train or adapt an appropriate model”, please?
Thanks
AV
We do have some updates here:
- NVIDIA Riva now supports German ASR. You can try it in your browser. It probably won’t perform well on medical terminology. With the English ASR model that we used in the demo, we trained a custom language model using KenLM and a bunch of medical-themed text pulled from PubMed and medical subreddits. Riva now has a variety of mechanisms to customize the recognition.
- Unfortunately we don’t have any German NER out of the box in Riva, let alone medical NER. You would need to train your own model and export it to Riva.
- GERNERMED looks like a nice model, though it won’t be possible to integrate it directly into Riva. Riva is intended for transformer-based models trained using TAO or NeMo. When I put up the link to GERNERMED, I wanted to highlight the training data they have shared, which looks like a translation of i2b2 data. You should be able to use this data to train a relevant NER model, using the instructions for TAO above.
I hope that helps!
Chris
I was looking to replicate this demo (https://developer.nvidia.com/blog/building-transcription-and-entity-recognition-apps-using-riva/) which led me to this forum entry. The solution link in this chat is trying to point to a “callcenter.html” demo. After some searching I realized the demo has been renamed to “Riva Contact” (Riva Contact — NVIDIA Riva). Hope this helps someone.
@svaha – Link fixed, thank you!