speechbrain/asr-transformer-transformerlm-librispeech · How to make inferences without transcription?

May 1

•

Hi,
I am trying to test the performance of the ASR using data without transcription. Unfortunately, the recipe requires the tokens_bos variable, which I do not have in my case. How can I make inferences to this model without having the transcription of the audio?

Thanks in advance for your help.
Best,
Matias

matiuste changed discussion status to closed May 1

matiuste

May 2

Hi,
I am trying to test the performance of the ASR using data without transcription. Unfortunately, the recipe requires the tokens_bos variable, which I do not have in my case. How can I make inferences to this model and be able to get the probability distribution over the predicted tokens without having the transcription of the audio?

I am using a speechbrain version 0.5.14 with a pretrained model.

Thanks in advance for your help.
Best,
Matias

matiuste changed discussion status to open May 2

matiuste

May 2

Solved: By giving an empty tensor.
tokens_bos = torch.tensor([[]])

matiuste changed discussion status to closed May 2