Closed Mark-Phillipson closed 3 months ago
I am curious, what prompt/use case was generating markdown for you? Have you updated the repository recently? I thought our system prompt is pretty strict at making sure there isn't markdown formatting unless it is requested.
Assuming this is an issue though, there are some heuristics that can be done to address what you say
Let me know if you have any other opinions about the text to speech functionality
Okay my bad I had the system prompt setup in settings without the markdown removal request.
The best use case so far with some programming code selected:
Model Explain To Speech
I'm not sure if this is possible but can you cancel once the speaking starts?
Sight free talon has a function called
actions.user.cancel_current_speaker()
Assuming you have it installed you can call this or define a new talon command to call it to stop speech.
Perhaps model stop speech
would be a good command to add. If you try this and feel is works feel free to submit a PR, should be quite simple
When the model is reading back the results to speech and it encounters the markdown characters for a heading or title (###), it explicitly says "hashtag hashtag hashtag." Can we address this by ignoring it or using another method?
Maybe get the model to return plaintext only?