Just started using ElevenLabs for a client's explainer video project, and I'm hitting a strange artifact. When I generate speech, especially with longer sentences, there's this faint metallic echo or reverb on certain consonants. It's most noticeable on "s" and "t" sounds. Almost sounds like the voice is in a small tin can for a split second.
I'm using the "Rachel" voice (Multilingual v2) on the "Stability" setting around 30% and "Clarity + Similarity Enhancement" turned on. My script is pretty standard marketing copy.
Has anyone else run into this? I'm trying to pinpoint if it's:
* A specific voice model issue
* Something with my audio output settings (though I don't hear it on other audio)
* A quirk of the "Clarity" setting
* Just a known artifact with the current generation
I love the platform for CRM voicemail drops, but this echo would be a deal-breaker for a polished video. Any workflow tweaks or alternative settings you've found to clean this up?
Spreadsheets > marketing slides.
I've heard that metallic edge on sibilants too, especially with some of the v2 voices. For me, it got better when I dropped the "Clarity + Similarity Enhancement" - that setting can sometimes over-process the high end. Try a generation with it off and see if the tinny echo persists.
Also, try bumping the stability up to 50-60% for longer sentences. Lower stability can make artifacts more pronounced as the model tries to be more expressive.
Curious if the issue is specific to longer audio chunks. Have you tried generating the script in smaller segments to compare?
Data is the new oil - but it's usually crude.