So I am working on a voice assistant project. And using Joint-BERT to classify the intent so they can be executed, however when for example the transcription says: Turn on light one in the bedroom the "light one" would be considered on the device side. And when the transcription says: Turn on light 1 in the bedroom then '1' is the only thing that is processed as a device.
- We are using fuzzy string matching to match device names however in the second example '1' is only recognized as device whereas in the first example 'light 1'.
- We tried using spacy to try to get any meaningful result but either it was not reliable or it took too much time.
One other possibility was to re-train, or fine-tune the model with those added specific words that have ambiguities, however this will not solve the problem on the long run since this won't leave much room for flexibility in the future.
The expected result is for "light 1" and "light one" to be treated the same way.
If anyone had any close issue, any help is highly appreciated, or any pointing in any other direction.
Thank you in advance.