It’s no secret that I use ElevenLabs speech synthesis for my stories. I’ve commented on it before, including creating a wish list of feature improvements. Today, I am sharing a couple of examples of a challenge and a simple (enough) workaround.
As I create audio files for Hemo Sapiens: Aftermath, I hear two problems with pronunciation. From a practical perspective, it costs me characters to re-do content. I am given 30,000 characters per month, so repeating passages can throw off my production schedule if I must wait for the next cycle for my characters to reset.
Hemo Sapiens
The first issue is that the text-to-speech engine arbitrarily flips back between the correct and incorrect pronunciation of the word hemo — /ˈhiːməʊ/ versus /ˈhɛməʊ/. In order to ensure it gets it one hundred per cent of the time—since it doesn’t support IPA—, I need to present it as heemo, thereby not only necessitating a re-do but adding a character along the way. To be fair, the IPA version would render nine characters, so there’s that.
Wounds that heal
As I listened to a passage with a homophone, wound, it pronounced the verb form as the noun form. As written, it looks like the top line. To force the correct pronunciation, I had to respell wound as wowned to shift from /wuːnd/ to /waʊnd/ .
- Jasmine exhales, releasing some of the tension that had her wound tight.
- Jasmine exhales, releasing some of the tension that had her wowned tight.
In my wish list video, I suggest a tagging scheme to remedy this. Of course, I’d hope the tags would not count against the character allotment.
Oh, well. First World problems, eh?
Ridley Park 