Enable manual sentiment for non-English language models that do not have built-in sentiment support

Manual sentiment may be used to augment the existing sentiment model, or to define sentiment rules for language models without a built-in sentiment model.

To enable manual sentiment, first create a directory on the V‑Blaze host for the associated acoustic model. This directory is typically used for models with built-in sentiment support. In this special case, it will remain empty, but it must be present for manual sentiment to function.

If adding this directory triggers an error message such as ERROR, sentiment classifier could not be properly initialized the error can be ignored; the manual sentiment process does not use the classifier.

The directory path varies by acoustic model and must follow this format:

/opt/voci/models/textproc/{ACOUSTIC_MODEL_CODE}/sentiment

For example, to enable manual sentiment for Spanish language models like spa-us:callcenter, the following directory must exist on the V‑Blaze host:

/opt/voci/models/textproc/spa/sentiment

Once this directory has been created, manual sentiment is enabled, but sentiment rules must be specified in the transcription request for any sentiment scoring to occur.

These rules may be submitted with the request using the senti_rules stream tag, or they may be referenced with the senti_list stream tag from a file saved on the V‑Blaze host in the following directory:

/opt/voci/state/sentiment