In this guide
For everyday mobile production
Automatically recognize spoken language and show it as captions in your live stream.
1. Live captions for mobile streams
LocalVocal is an OBS extension for local speech recognition and captions. It lets viewers follow speech better even with the sound off. Depending on your configuration, translation and output to text sources are also possible.
In an IRL setup, recognition runs wherever OBS runs. The microphone signal arrives in OBS, the filter turns it into text, and a text source overlays it on the picture.
2. Requirements
| Requirement | Why it matters |
|---|---|
| Compatible OBS and plugin version | Otherwise the plugin can't be installed or started cleanly |
| CPU, GPU, and RAM headroom | Recognition runs on the OBS PC and puts extra load on it |
| Clean microphone source | With wind and noise, recognition makes more mistakes |
| Storage space | The model and temporary data take up space |
3. Setup
- Check the release and installation notes in the official project.
- Close OBS, install the matching version, and start OBS.
- Add the LocalVocal filter to your clean microphone source.
- Choose the language and a model that fits your PC's performance.
- Connect the text output to an easy-to-read OBS text source.
- Test delay, line length, position, and contrast with movement on screen.
Which options the filter has depends on your installed version. What applies is in the official project.
4. Accessible captions
| Feature | Recommendation |
|---|---|
| Font | Large, with strong contrast and a calm background |
| Length | Two short lines at most |
| Position | Don't place them over important parts of the picture or platform overlays |
| Labeling | Add a note that the captions are generated automatically and may contain mistakes |
5. Privacy
Local processing can avoid sending spoken content to an external speech recognition service. Transcripts are still created, though, and they may contain personal or sensitive statements. Only keep text files, logs, and recordings as long as necessary, and review them before publishing.
Keep in mind that in an IRL stream, conversations with passersby end up in the microphone too. More on this is in the IRL4YOU privacy policy (German).
6. Testing before the stream
- Watch the load and temperature of the OBS PC while it runs.
- Check how far the captions lag behind the audio.
- Check that you can read the captions clearly with a moving picture and changing brightness.
- Test recognition in wind and ambient noise, ideally with a private test recording.
7. Common questions about LocalVocal
Does speech recognition run on the smartphone or encoder?
Usually not. It runs on the OBS PC, not on the mobile encoder. Make sure that PC has enough headroom in CPU, GPU, and RAM, and test load and temperature before you go live.
Is my speech data sent to a service?
With local processing, spoken content doesn't have to go to an external speech recognition service. Check the official project to see whether that holds for your settings. Transcripts are still created, and they can contain personal or sensitive statements.
How good are automatic captions?
They make mistakes, especially with wind and ambient noise. Use a clean microphone source and add a note that the captions are generated automatically. You'll find tips for audio on the go under IRL audio and Microphones for IRL streams.
Which model should I choose?
One that fits your PC. A larger model can need more resources. Start with a smaller one, watch load, temperature, and delay, and then move up step by step.
Where do I get LocalVocal and how do I install it?
Releases and installation notes are in the official project on GitHub. Pick a plugin version that matches your OBS. Close OBS before installing and restart it afterward. You'll find OBS basics in the OBS guide.
Setup service & consulting
If you want to build the extension into your OBS and IRL workflow, you can get in touch on the IRL4YOU Discord.