Possible outputs?

#1
by Pranavan - opened

What could be the possible output/tags?

For example:
Speech - What's the weather like
SLU Results - weatherquery ▁what s ▁the ▁weather ▁like

What are the words/tags that the model has been trained on? (similar to 'weatherquery' on the example above!)

ESPnet org

Thanks for your interest!

The model has been trained on SLU dataset SLURP (https://arxiv.org/abs/2011.13205). The dataset has 18 scenarios (like weather in the given example) and 46 actions (like query in the given example). Kindly refer to the dataset repo (https://github.com/pswietojanski/slurp) to know the exact tags the model has been trained on.

The model is then trained using an auxiliary ASR objective by making the model first generate intent followed by ASR transcript. Please refer to our paper(https://arxiv.org/abs/2111.14706) for further details.

Let me know if you have any questions!

Pranavan changed discussion status to closed

Sign up or log in to comment