Question

You are building a hands-free kiosk app in the Foundry portal. Users will speak a question out loud, and the app must reply with a spoken answer generated by a deployed multimodal model that can accept audio input directly. Which approach best implements this using a single deployed multimodal model plus Azure Speech in Foundry Tools?

Accepted Answer

Send the user's audio to the multimodal model to generate a text response, then use Azure Speech text-to-speech to convert the response into spoken audio for playback.

Answer

Use Azure Speech text-to-speech to transcribe the spoken question, then send the transcript to a regression model for a numeric answer.

Answer

Send the audio directly to an image-generation model, which will return a synthesized voice clip as its output.

Answer

Use a translation model to convert the spoken question to text and back to audio without involving the multimodal model.