Trying to use Fliki for a product demo video. The AI narration keeps butchering key brand and tool names. It's breaking the flow.
Examples:
* "Terraform" pronounced like "Terra-form" with equal emphasis, sounds robotic.
* "Kubernetes" is hit or miss, often completely wrong.
* "NGINX" comes out as "N-Ginx" letter-by-letter.
* My company name got mangled.
I've tried adding phonetic spellings in the script text, but results are inconsistent. The workaround of splitting the word and recording a custom voice clip for each one is not scalable.
Is there a better method, or is this a known limitation with the TTS engines they use?
—cp
Ugh, yes, this is a huge pain point with TTS for technical content. I've hit the same wall.
The phonetic spelling trick can sometimes work if you force a specific SSML tag, but it's totally engine-dependent. For example, some engines will respect a `` tag if the API supports it, but most consumer tools like Fliki abstract that away.
Curious - have you tried replacing "Kubernetes" with "Kube" in the script? It's not ideal, but for a demo flow it might be less jarring than the wrong pronunciation. For NGINX, I've had to write it as "Engine-X" before, which feels wrong but the client didn't notice.
I think this is less a Fliki-specific issue and more a general TTS training data problem. These engines aren't trained on enough DevOps meeting recordings!
Webhooks or bust.