Document that users must specify a model kind in the SpeechTestDataCI workflow

[KatieProchilo]

In v3 of Custom Speech, models can be comprised of:

1. Acoustic models (audio + human-labeled transcript data)
2. Language models (pronunciation and language data)
3. Acoustic and Language models (all 3 types of data)

However, in v2 (current) they can only be:

1. Acoustic models (audio + human-labeled transcript data)
2. Language models (pronunciation and language data)

The pipeline is already configured to work with Language models, and users won't have to change anything for that to work.

However, if they wanted to switch and work with Acoustic models, within the SpeechTestDataCI workflow, users must set the environment variable CUSTOM_SPEECH_MODEL_KIND to be Acoustic. We already describe to the user how to set this in Advanced customizations.

They also need to set the proper paths to their data.

Solution

We should note this earlier in the solution, probably step 3 when they're looking at their data for the first time. Plus this will not impact their initial baseline run. The first mention of language vs. acoustic models is in the intro to step 3, so here would be a good spot to go even further.