A few years back, we launched an app called SpeakChat for voice messaging. The magic of it was that it combined voice message with speech-to-text — and it worked in multiple languages. The user could speak or type their message and on the other side, the recipient could either read the message as text or listen to it.
This provided the option to decipher the message if the speech transcription was incorrect. The backup meant that there was always a way to interpret the message.
The Echo or Google Home could benefit from this feature. “Alexa, what did you think I asked you?” or “OK Google, repeat back what I said” can allow the system to learn from errors.