How is a "phrase list" (from AI Speech) different from a "keyword list"? (Copilot, AI Search) or "phrase dictionary" (AI Translator)?

Question

Hi,

I`m wondering about differences and similarities - and the reasons.

AI Speech: A phrase list is a list of words or phrases provided ahead of time to help improve their recognition. Adding a phrase to a phrase list increases its importance, thus making it more likely to be recognized. Punctuation is ignored (...makes sense...)

Copilot: Shorter trigger phrases are better, and you should aim for 2 to 10 words per phrase.

AI Translator: Dictionary entries are case- and punctuation-sensitive... (...also makes sense HERE). The phrase dictionary should be used sparingly. When a phrase within a sentence is replaced, the context of that sentence is lost or limited for translating the rest of the sentence

Is the underlying technology similar or is it just a case of coevolution?

Answer 1

@Michael Schmidt With respect to AI speech and AI translator I think the context is similar which is the service tries to utilize these list to help in better speech recognition and for translator it tries to customize the translation for the words or phrases that you have added.

In both cases, they come under customization scenario but with Speech, it is a lightweight way to customize some text recognition for words or phrases that we expect to be mispronounced by the standard model. For scenarios, that involve a lot of data i.e words or phrases using a custom model is required but that needs training a model on top of base models and deploying it.

With the translator the phrase dictionary is applied to translation for a customization scenario where the output and input are both text data. The scenario is the same as phrase list of speech but applied with translation context.

I am not too familiar with copilot list, could you add a link to the documentation you have quoted for copilot?

I hope the above information is helpful.

If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.