Twitch added mandatory captions for accessibility in late 2025. YouTube Live now auto-promotes streams with captions in search. Facebook Gaming has a "captions first" promotion program. The math is simple: live captions get you more viewers, better discoverability, and protection from accessibility lawsuits.
The good news: 2026 is the year live captions got easy. You don't need a paid service or a complex setup. The free LocalVocal OBS plugin runs Whisper locally on your PC and delivers captions with 2-3 seconds of latency โ fast enough for live conversation.
| Tool | Cost | Latency | Privacy | Best for |
|---|---|---|---|---|
| LocalVocal (OBS plugin) | Free | 2-3s | Local | Solo streamers |
| StreamLive Captioner | $15/mo | 1-2s | Cloud | Pro studios |
| Built-in Twitch/YouTube auto-captions | Free | 8-15s | Cloud | Backup only |
This guide focuses on LocalVocal because it's free, fast, and private. If you need professional studio latency, look at StreamLive. The built-in auto-captions are too slow to be useful for live conversation.
If you don't already have OBS, download it from obsproject.com. OBS 30 added native support for AI plugins, which LocalVocal relies on.
Go to the LocalVocal releases page and download the installer for your platform:
obs-localvocal-X.X.X-windows-installer.exeobs-localvocal-X.X.X-macos-x86_64.pkgobs-localvocal-X.X.X-macos-arm64.pkgRun the installer. Restart OBS if it was open.
In OBS, go to Docks โ obs-localvocal (or Tools โ LocalVocal depending on your version). A new dock appears.
Click Settings:
ggml-base.en for English (fast) or
ggml-large-v3 for accuracy (slower). The first time you select a
model, LocalVocal downloads it (~500MB for base, ~3GB for large).en for English or your target language.In OBS's Sources panel, click + and add LocalVocal Captions. A text source appears showing live captions.
Customize the appearance in the source properties:
Talk into your microphone for 5-10 seconds. You should see captions appear within 2-3 seconds. If you see nothing:
LocalVocal can run on CPU only, but a CUDA GPU (NVIDIA) or Metal GPU (Apple Silicon) makes the large model usable in real-time. On an M1 Mac or GTX 1060+, you can run large-v3 with 2s latency. On a Chromebook or old laptop, stick with base model and accept 3-4s latency.
Background noise kills Whisper accuracy. If you're streaming from a noisy environment, add OBS noise suppression (RNNoise works great) before LocalVocal's audio source. LocalVocal doesn't do noise reduction itself.
Whisper doesn't know your streamer's name or your game's terminology. Use LocalVocal's Translate and Replace feature to substitute common mistakes. For example, if you play a game called "Dragnor" and Whisper keeps writing "Dragner," add a substitution rule.
If you need higher accuracy or translation, the next level is cloud captioning:
For most streamers, LocalVocal + free is plenty. Cloud options are useful for multi-language streams (Twitch viewers in 50+ countries) or for compliance captions where 99% accuracy matters.
Live captions solve the accessibility problem, but they also create new opportunities:
Yes, if you enable Translate to English in settings. For non-English target languages, you need to chain LocalVocal โ translation API โ captions.
Yes, but you may need to resize the captions source. Position captions higher (around 60% from top) so they don't overlap with TikTok's UI.
Base model: 20-40% of one CPU core on a modern laptop. Large model: 60-100% of one core + 2GB GPU memory. Plan accordingly.
Download LocalVocal, plug it into OBS, and you have live captions in 10 minutes. The accessibility benefits alone are worth it โ and the discoverability boost from captions is a nice bonus.