Repository navigation
[Discussion] Voice support #653
Description
Goal of this issue is to start a discussion on adding voice prompting in T3code.
I've worked on a demo of voice support.
A simple button to start recording + associated config.
You can out the code here: abaudhuin#1
recording_20260309-111328.mp4
My current thing has important compromises, just a starting point for the discussion.
It uses a local Parakeet 0.6B v2 model running on CPU https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2. Which comes with its rust sidecar...
But it works.
Is voice something you are considering?
I'm willing to contribute on this topic on whatever your vision would be.
Activity
aaditagrawal commentedon Mar 9, 2026
I think most people already use third-party apps for voice TTS, but this could be a good addition.
But for those who already have it, it's also duplication, both in compute and memory.
I would prefer if this would be an opt-in feature.
abaudhuin commentedon Mar 9, 2026
Clearly, it should be opt-in.
The question of using a local model is also questionable (this one is multi-platform, multi language).
Could be using one/multiple cloud provider for the voice model.
[-]Voice support[/-][+][Discussion] Voice support[/+]maria-rcks commentedon Mar 15, 2026
out of scope, there are apps that do this already.
gpu004 commentedon Apr 14, 2026
openai provides there speech to text model usage in there subscriptions, an voice input can be made, would be an awesome quality of life feature
akhmerov commentedon Aug 7, 2026
codex app-server now supports voice input (as of version 0.147); is that perhaps a reason to revisit this issue?
fix(provider): harden Antigravity adapter (pingdotgg#653/pingdotgg#696–…
7heMech commentedon Sep 10, 2026
Both antigravity and codex support voice inputs, antigravity is insanely good at it with their new 3.5 transcribe model.
bdbch commentedon Oct 2, 2026
I would just use a standalone application for this honestly because then you don't have to build it into all your apps and you can just use it everywhere. For example I use OpenWhispr and it works really well.