From 665279f7cd1cc2c0b7ff77c9af55ac2437693e39 Mon Sep 17 00:00:00 2001 From: Pinapelz Date: Mon, 21 Sep 2026 20:05:11 -0700 Subject: feat: move to virtual sink and loopback audio device as default method - creates a virtual audio sink on startup and a loopback - enables listening on loopback - provide finegrain control over what you want the AI to hear --- README.md | 20 +++++++++++++++++++- 1 file changed, 19 insertions(+), 1 deletion(-) (limited to 'README.md') diff --git a/README.md b/README.md index e455194..55b5ae5 100644 --- a/README.md +++ b/README.md @@ -13,7 +13,25 @@ uv sync uv run server.py ``` -`server.py` serves a backend for translating incoming audio data. It expects some other client to hit the `/events` endpoint to fetch the translated data. A GUI is available for configuration +`server.py` serves a backend for translating incoming audio data. It expects some other client to hit the `/events` endpoint to fetch the translated data. A GUI is available for configuration. + +## Linux virtual audio sink/input (PipeWire) +On Linux, startup now attempts to create: +- a virtual sink via `module-null-sink` + - sink name: `auto_live_tl_sink` + - sink description: `auto-live-tl-virtual-sink` +- a virtual input source via `module-remap-source` + - source name: `auto_live_tl_input` + - source description: `auto-live-tl-virtual-input` +- a headphone/speaker mirror loop via `module-loopback` + - source: `auto_live_tl_sink.monitor` + - sink: your current default output sink + +This gives you a direct mic-like input device in the Settings UI (shown with `[virtual input]`) while still hearing audio on your normal output device. + +If your PortAudio backend cannot see PipeWire sources, the app adds a fallback option named `auto-live-tl-virtual-input (PipeWire direct)` and captures from PipeWire using `ffmpeg`. + +If `pactl` is unavailable (or device creation fails), the app continues running and falls back to normal input devices. # Whisper + Ollama (Local Setup) > It's highly recommended that you run this with a GPU, running with CPU is possible but inference will be very slow outside of using tiny models (which compromise accuracy) -- cgit v1.2.3