Hi folks! I've been experimenting with WellSaid for a few weeks now, mostly for generating narration for internal data pipeline tutorials. I remember being completely overwhelmed by the voice options at first, so I feel your pain!
For a complete newbie, I'd suggest a structured approach:
* **Define your "use case pipeline":** Think of your script as data flowing through a pipeline. What's the destination? A customer-facing explainer, internal training, an audiobook chapter? The tone (authoritative, friendly, conversational) is your first filter.
* **Start with the "Featured" voices:** WellSaid does a good job curating these. They're generally the most versatile and reliable. Pick 2-3 that seem to match your tone and run the same short script through all of them. Listen to the outputs back-to-back.
* **Pay attention to data points:** Don't just listen for "likability." Note:
* Clarity on technical terms or specific jargon you use.
* Pacing and where natural emphasis falls.
* How it handles long sentences versus short ones.
My personal workflow is to treat it like an A/B test. I'll often generate a key paragraph with two different voices, then get quick feedback from a colleague. It's surprising how much preference can vary.
Has anyone else developed a systematic way to evaluate or shortlist voices for different project types? I'm curious if you prioritize consistency across projects or choose the absolute best fit for each one individually.
—Claire
That A/B test approach is spot on, especially for internal tutorials. I'd add one thing: listen on the actual device or platform where your team will consume it. A voice that sounds great through my studio headphones can sound thin on a laptop speaker in a busy office.
I also create a small rubric for scoring, like a 1-5 on clarity for acronyms and pacing. It forces me to be objective instead of just picking the voice I like.
terraform and chill
Treat it like an A/B test is perfect advice. I'd push that a step further and actually use an A/B testing platform for the feedback collection, even internally.
Set up a quick survey in something like Typeform or even a Google Form, embed the audio samples, and ask your team to rate them on clarity and tone without knowing which voice is which. It removes bias and gives you real data.
I've found that the voice I *think* is most authoritative isn't always the one my team absorbs information from best. The data usually surprises me.
Cheers, Henry