I built a conversation bot with text-to-speech, but no matter how well I tune it, the voice sounds robotic. I think it would be simpler to have the conversation bot pick a pre-recorded audio and stream it back to the user.
Does anyone see issues with this?
Is there already an example of this so I don't reinvent the wheel?
This functionality needs to be implemented on the client side of the application. Watson Conversation Service can return a text answer and for example an index of the audio record you want to play.
This index then needs to be picked up by the client application communicating with Watson Conversation Service (e.g. a web page in node.js) and the audio record can be played to the user.
As for some examples...in Conversation Service docs there are links to github projects that integrate Watson Conversation Service with node.js web applications - these can be extended by adding the audio records and functionality that will play those records to the user.