[proposal] Add Speech to Text in Kilo for the Chat

[adamhill]

Challenge

We currently have no way to talk to Kilo to interact with the LLM.

This may seem like a novelty to some people (hands off vibe coding) but its needed for developers that have no mobility in their hands, but otherwise have normal speech abilities.

We have a long weekend in the states so feels like a nice time to dig in and start

Proposed Solution (high level)

• Add a button to the chat widget to toggle the microphone off and on
• Pipe the recognized text into the chat box
• After a timeout, simulate pressing enter

Basically, mimic the behavior of the Copilot VS Speech feature

Implementaton Details

• It will reuse the Speech Provider API's already existing in VSCode & the VS Speech Extension that Copilot uses.
• These API's are still in the .proposed.d.ts namespace
• But...there are rules around .proposed API's
  • We cannot release an Extension in the Marketplace with .proposed APIs & we have to use Insiders to test it
  • So... back in April I made an Issue in the vscode repo to move them out of .proposed and into the released API or get a headless or not UI widget that could recieve text from the VS Speech Extension
  • It got 20 votes and now it is in the Backlog. Yay!
• Now we wait for the API to move 💤
• But I want to get it done and show it to the vscode team to convince them to move it up the Backlog or just approve migration.

I have pinged (@bpasero) who seems to have implemented various VS Speech bitsand is its "shephard" according to commits, comments and external repos.

After Thursday, this might be much easier since they have open-sourced copilot-chat

Alternative Implementations considered

Someone made a furtive collab over in the Roo Discord to try and do a web socket version that has not gone anywhere

I looked at sherpa-onnx it looks doable but waaaaaay more work

Additional Context & Mockups

I have a bunch of reasearch in Obsidian, but nothing thats easy to share. I will see what PDF export is like in Obsidian (ironically a Markdown first based system)

I am willing to see this through implementation and rebasing as needed as we wait for MS to release the API's to stable

Thanks.

[markijbema]

Get it done and rebase until it lands in vscode sounds like a great plan to me. You can even use the same tricks we use ourselves for the roo merge (mark changed areas, so the merges are easier to understand)