Hosted Inference API with SpeechBrain returns arror

Models
1

I’m using the SpeechBrain toolkit to fine-tune XLSR. Everything worked well but now when I’m trying to use the Inference API, I’m getting these error :

Can't load feature extractor for 'facebook/wav2vec2-large-xlsr-53'. Make sure that: - 'facebook/wav2vec2-large-xlsr-53' is a correct model identifier listed on 'https://huggingface.co/models' - or 'facebook/wav2vec2-large-xlsr-53' is the correct path to a directory containing a preprocessor_config.json file

I observe a similar result with the SpeechBrain models on HuggingFace that use XLSR. So I guess that the problem is not only with my implementation but probably related to the Hosted Inference API.

I use to present the results of my works using these graphical interface, so any thoughts to solve this issue will be very welcome!

2

Hello,

The task itself (sort of masked speech modelling) is not defined for the inference API. The widget also doesn’t work at the moment. If you fine-tuned it on a downstream task that exists, the inference API should work on that model. Is there any link to your model?

3

Hello,

Thank you for your fast response. Here is one of my models : https://huggingface.co/nairaxo/dvoice-swahili. It was inspired by this SpeechBrain recipe.

4

@nairaxo is there any problem with your model on inference API?

5

Yeah. This is what I get when I try to infer the model using the inference API. It works usually, but since 3/4 days, it not working. I don’t know if it’s related to the “facebook/…” models that I fine-tune or to the inference API widget as you say.
This what I got instead of the speech transcriptions…

inference
inference899×1599 100 KB

1 Like
6

Pinging @anton-l here :slightly_smiling_face:

7

We had an issue with the Inference API, sorry for the bug.

It should be up again!

1 Like
8

Thank you all. It works now! :+1:t6: