This guide walks through how to capture live programme audio, generate multilingual AI captions through KUDO, overlay them onto a video feed in OBS Studio, and stream the final output to Vimeo through an Epiphan Pearl encoder.
Audience: Hosts, Producers, Operators, and AV technicians configuring AI caption overlay workflows for live productions.
How the Workflow Fits Together
The workflow runs across four stages. First, a dedicated ingest computer joins the KUDO meeting as a Participant and routes NDI programme audio into KUDO AI for caption generation. KUDO processes that audio and delivers real-time captions through the KUDO Mobile Link. OBS Studio then composites those captions over the programme video feed. Finally, an Epiphan Pearl encoder takes the OBS output and streams the captioned programme to Vimeo via RTMP.
Before You Start
Confirm that the KUDO meeting has been created with AI Captions enabled and the Mobile Link is available for the desired caption language. Vimeo streaming credentials should be on hand, the TriCaster programme output should be accessible via NDI, and the following software should be installed before beginning: NDI Tools, VB-Audio Virtual Cable, OBS Studio, the DistroAV OBS plugin, and Google Chrome or Microsoft Edge. The Epiphan Pearl encoder should also be accessible on the local network.
For the best production stability, use wired Ethernet connections on all systems, separate the ingest and overlay workflows across different machines when possible, and test NDI routing and caption latency before the live event. Validate the Vimeo ingest before going live.
Important: Audio quality, network stability, and source platform performance directly affect AI caption quality and synchronization.
Equipment
You'll need a TriCaster as the programme video and audio source, an Epiphan Pearl as the RTMP encoder for Vimeo delivery, a dedicated Ingest PC to route programme audio into KUDO, and an Overlay/Production PC running OBS Studio. The Ingest PC and Overlay PC can be combined into a single machine if CPU resources allow. Always test under production load conditions before going live.
Phase 1 — Configure the Ingest PC
The goal here is to capture programme audio from the TriCaster NDI output and send it into KUDO as the microphone source. KUDO AI then generates live captions from that incoming audio.
1. Install NDI Tools and VB-Audio Virtual Cable, then restart the computer.
2. Open NDI Virtual Input and select the TriCaster programme NDI source. Confirm the NewTek NDI Audio device appears in Windows audio settings and that audio signal activity is visible. For additional browser compatibility, optionally route the signal through VB-Cable as the browser microphone input.
3. On the Ingest PC, open Chrome or Edge and join the KUDO Participant/Speaker link as a Participant. Inside KUDO, select the NDI audio device or VB-Cable output as the microphone source, confirm microphone activity is visible, and mute local speakers to avoid feedback.
Phase 2 — Retrieve KUDO Captions
4. Open the KUDO Mobile Link for the desired caption language in a browser. This caption feed will be embedded into OBS Studio as a Browser Source in the next phase.
Phase 3 — Configure the Caption Overlay in OBS
The goal here is to composite the KUDO caption feed over the programme video, creating a single captioned output for Vimeo delivery.
5. Inside OBS, add a new NDI Source and select the TriCaster programme feed. Set the output resolution to match your production format. The programme feed should appear in the OBS canvas.
6. Add a new Browser Source and paste the KUDO Mobile Link URL. Set the resolution to match the production output — 1920x1080 is recommended.
7. Move the caption layer above the programme video source and position captions in the lower third of the canvas. Resize or crop the Browser Source as needed. Adding a semi-transparent black background behind the captions can improve readability. Preview the scene and confirm captions appear correctly over the programme feed.
Phase 4 — Configure OBS Output to Epiphan Pearl
8. Choose one of the following methods to deliver the OBS output to the Epiphan Pearl:
| Method A: NDI Output (Recommended) | Method B: HDMI Loop-back | |
|---|---|---|
| How it works | OBS sends its output as an NDI stream over the network; Pearl ingests it as an NDI source | OBS Virtual Camera output is captured through a video capture card or direct HDMI connection into the Pearl |
| Requirements | Pearl-2 or Pearl Nexus with NDI input support; DistroAV OBS plugin installed | Overlay PC with HDMI output connected directly to the Pearl HDMI input |
| OBS setting | Go to Tools > NDI Output Settings and enable Main Output | Go to Tools > Start Virtual Camera, then connect HDMI to the Pearl |
| Pearl setting | Add an NDI source and select the OBS NDI output as the channel input | Set the Pearl channel input to HDMI IN connected to the Overlay PC |
Phase 5 — Stream to Vimeo
9. Inside Vimeo, open the live event and locate the RTMP URL and Stream Key. Keep these available for the next step.
10. Inside the Epiphan Pearl web interface, select the correct input source and add a new RTMP destination. Paste the Vimeo RTMP URL and Stream Key, then configure encoding settings. Recommended output: H.264 video, AAC audio, 1080p, 25 or 30 fps. Click Apply, then Start Streaming.
Troubleshooting
| Issue | Possible Cause | Recommended Action |
|---|---|---|
| No captions appear in OBS | Incorrect Mobile Link URL | Verify the correct Mobile Link was added as a Browser Source |
| Captions are delayed or inconsistent | Poor source audio quality | Check programme audio quality and microphone routing |
| OBS Browser Source appears blank | Browser rendering issue | Refresh the Browser Source or reopen OBS |
| Audio feedback loop occurs | Local monitoring enabled | Mute local speakers or use headphones |
| Epiphan Pearl doesn't receive OBS output | NDI or HDMI routing issue | Verify OBS output settings and Pearl input configuration |
| Vimeo stream fails to start | Invalid RTMP credentials | Verify the Vimeo RTMP URL and Stream Key |
| NDI audio not detected in KUDO | Incorrect audio routing | Verify NDI Virtual Input and VB-Cable device routing |
| Caption overlay blocks video content | Overlay positioning issue | Reposition or resize the Browser Source in OBS |
Known Limitations
AI caption quality depends heavily on source audio quality. Network instability may affect NDI routing and caption synchronization. OBS Browser Sources may need to be refreshed during extended productions, and combined ingest and overlay workflows on a single machine may require additional CPU resources. Caption timing may vary slightly depending on AI processing latency, and Vimeo synchronization behavior can vary by network conditions.