I've been testing Murf for generating voiceovers for internal security training modules. The technical content is fine, but the AI's handling of proper nouns—especially vendor names, tool names, and internal project codenames—is embarrassingly bad. It's turning "Snyk" into "snick" and "Istio" into "is-tee-oh." This isn't a minor quirk; it breaks credibility in a training context.
I've tried the obvious:
* Adding phonetic spellings in the script (e.g., "Snyk (pronounced 'sneak')").
* Using the pronunciation guide in the studio UI, which feels like a workaround for a handful of terms.
* Manually editing the audio, which defeats the purpose of automation.
What I'm not seeing is a scalable, programmatic approach. For an enterprise rollout, I can't manually correct dozens of unique proper nouns per project.
Has anyone found a method that doesn't involve:
* Pre-recording a library of corrected clips (maintenance nightmare).
* Accepting butchered pronunciations (unprofessional).
* Abandoning the tool for anything with jargon (not an option).
Specifically:
* Is there a way to bulk-upload a custom pronunciation dictionary via API?
* Does the "voice styling" feature have any hidden controls for this?
* Are certain voices or engines (like their "AI Engine v2") better at this than others?
If the platform can't handle consistent technical terminology, its utility for anything beyond generic marketing copy is severely limited. Postmortems from anyone who's tried to scale this would be appreciated.
- Nina
- Nina