A-133For School Admins and Directors
Workstation narration: your own GPU PC narrates a course's lessons and sends the audio here
A switch and a status block in a master course's Narration panel. Your GPU workstation narrates lessons with the free Kokoro voice about 50 times faster than the server and sends the finished audio; approved lessons are never touched.
Open this in VocationalLMSAI Course Builder > a course > Narration for this course > Workstation narration
Who can do this: Only the VocationalLMS Super Admin can see or change the switch. The PC worker signs in with a private token saved on the server and on your PC; it can reach only these narration addresses, and only VocationalLMS master courses unless you list a school on the server. Students, School Admins, Directors, Instructors and auditors are refused.
What it is for
The server records Kokoro narration at about the speed of speech, so an 11-minute lesson takes about 11 minutes. Your workstation (the RTX 3090 PC) records the same lesson with the same model and the same voice in about 12 seconds. Workstation narration lets that PC pull the lessons that still need narration, record them, and send the audio back. The server stores each clip exactly where its own narration would be, so the slide pictures, the video, the captions and the protected streaming are built the usual way afterwards. It costs nothing: no voice account and no picture service is used. A lesson is never recorded at ElevenLabs by this route, and the server refuses to serve a course that is not set to Kokoro.
Steps
Get the course ready
- Open the master course page and click Narration for this course.
- Under Engine choose Kokoro (local, free). This is required: it is what makes sure no missing slide can ever be recorded at ElevenLabs.
- Under Workstation narration check that Token on the server says set, then tick Allow the workstation to send narration. It is off until you tick it.
Run the worker on the PC
- On the workstation run
py tools/kokoro_pc_worker.py --server https://vocationallms.com --course-id 9with the Kokoro environment (C:/ai-local/venv-tts). Add--max-lessons 3for a first try. - It pulls a batch, records every slide on the GPU, sends each clip, then finishes the lesson. Leave it running; if the site is being updated or the internet drops it keeps the finished clips and sends them later.
- To stop it at any time create a file named STOP in its work folder (or press Ctrl+C). It stops between slides and carries on from where it was the next time.
Check what arrived
- The block shows how many lessons still need narration, how many are held by the worker right now and the last clips received.
- A received lesson becomes voiced. To build its video, tick nothing here: run the worker with
--render(one video at a time, in the quiet hours) or use the normal Build video button on the lesson.
What happens next
The lesson is voiced with the same fields as any recording. Building its video is the heavy step, so it runs one lesson at a time on the server; plan it for the evening deploy window.
Common problems
| # | Problem | What to do |
|---|---|---|
| 1 | The worker says the switch is off (403). | Tick Allow the workstation to send narration in the course's Narration panel. |
| 2 | The server says to set the course to Kokoro first. | Choose Kokoro under Engine in the same panel. The route refuses any course that could fall back to ElevenLabs. |
| 3 | A lesson never appears in the worker's list. | It is approved or in review, it is being written or rendered, a job is running on it, another worker holds it, or every slide already has narration. |
| 4 | The server answered that the text changed. | Someone edited a slide after the worker pulled it. The worker throws that lesson away and pulls it again; nothing wrong was stored. |
| 5 | I see two voices in one lesson. | A slide that already had a recording for its current text is never replaced. Edit the slide's text, or ask for the lesson to be re-recorded, to change it. |
