WIP: local text-to-speech, speech-to-text in VAM, and more (Patreon)

TTS & STT

This was a quick first test I did for speech to text implemented in vam. I built a tool named SPQR Text Audio Tool that I'll release separately that helps with local TTS voices and with speech to text. Initially I built it just as a way to play offline voices, but I ended up adding speech to text too today. In the video I had my mic set up so that it captures what's playing on the pc.

I also added ElevenLabs AI voices support earlier: https://www.youtube.com/watch?v=-vmudQFXYy0 

I didn't show it yet but I also added support for RiveScript, which is an old-school AI chatbot language, similar to AIML, which allows classic chatbot functionality. I also extended some stuff to make it a bit more powerful for the use-case of Alive. This allows creating custom commands that add more practical use and smarter narrative than what's possible with plain & small AI models:

LOCAL ASSISTANT

When activated, Alive will pass the chat message first to the local assistant and tries to identify if it's something it can cover. If it can't, it passes it over to the AI. In a sense the local assistant aims to act as the logical brain (more reliable, predictable, with specialized functions), while the AI is the creative brain.

The goal is to allow people to create their own commands and mod the local assistant, which should in theory work as a way to mod/drive the AI itself (the AI will 'think' that what the local assistant wrote is what the AI wrote earlier, so it will incorporate & build upon that information).

I only did two very quick commands for it to test it:

The first is a debugging command to test that the system I built works asynchronously (the local assistant is able to wait for stuff before replying). This could be used to make the AI able to get data from the internet, like RSS news, weather, look up for stuff.

The second is a basic calculator example. When writing something like "Calculate (2+1)/69", the local assistant will pick that up and calculate it, rather than ask the AI to do it.