These type of algorithms are only able to match audio when the audio was added to an index beforehand. So it only capable to match pre-recorded audio. A live rendition of a text or song would not work since it would not match the index.

ESP32 has a variant which is compatible with ESP-ADF (audio development framework)

I haven't checked them in a while but they could've do on-device activation and text conversion before sending to Alexa API.

The problems were the docs are mostly chineese and its in early stages.

It would work if you can sing out the phrase ;). Basically a real life version of a musical.