Use cases
For e-learning and training teams
The course will change. That is the argument for synthetic narration.
Last checked 2026-08-14
In short
Training teams narrate one line per slide, keep one voice and one pronunciation list across the whole course, translate the same CSV into any of 35 languages, and ship an SRT with every module. When a policy changes, three lines are re-rendered in the identical voice - which no studio session can do a year later.
What the work looks like
- A CSV per module: slide id, script, voice. It is the audio source of truth and it survives staff turnover.
- A workspace pronunciation list for product, policy and regulator names.
- Rendered as a queued job, one file per slide, named by slide.
- Translated versions from the same CSV, same mapping.
- Captions from the same render, because they are usually mandatory.
Procurement's questions, answered in advance
Who processes the data and where (Scalingo, Paris, with the full subprocessor list published), what is stored and for how long, how to export or delete a workspace with one click, and whether the audio is marked as synthetic (yes - AudioSeal, plus a provenance record). The privacy, security, DPA and subprocessor pages exist so you can send links instead of writing emails.
Keeping a course maintainable for three years
The reason to narrate synthetically is not the first version, it is the seventh. That only works if the audio can be reconstructed without you, so treat the CSV, the voice slug and the pronunciation list as course assets and store them with the SCORM package.
- One row per slide, keyed by the slide id the authoring tool uses - not by order, which changes.
- Voice slug recorded in the file, so a colleague in 2029 gets the same narrator.
- Pronunciation list at workspace level, so every module and every language version inherits it.
- Render as a job, one file per slide, named by slide id.
- When the policy changes, edit the rows that changed and re-render those.
A studio session cannot do step five a year later at any price, which is the entire argument for doing it this way.
Where it falls short
Role-play scenarios with two characters in conversation still sound synthetic, and a course whose subject is empathy should be read by a person. Everything procedural is fine.
Questions
- Can we update one slide's narration next quarter?
- Yes - change the row, re-render that line, identical voice.
- How many languages?
- 35 for translation, including 23 Indian languages.
- Do you sign a DPA?
- The processing terms are published and the subprocessor list is public; write to us for a signed copy.