A cider-press room runs both a Riva booth and a pour-shot generator. An intern pastes the booth’s results into the same “quality” column as the pictures. What should happen?
Select an answer to reveal the explanation.
Short Explanation
Lumping a spoken booth and a pour photo into one “quality” cell is like scoring soup and stained glass with the same ruler. Speech needs transcription, task success, or listener ratings. Pictures keep their own scores.
Full Explanation
Multimodal programs must match metrics to the modality under test. Riva pipeline success uses speech-appropriate measures; image generators use image metrics. Collapsing both into one quality column hides real experimental signal and confuses go/no-go decisions.