Just got my monthly invoice from Resemble and did a double-take. My usage felt consistent with last month, but the total was nearly three times higher. After digging into the billing details, it looks like everything got moved to the new "Pro Voice" tier pricing.
I've been using their API for generating short voice clips for automated system alerts. Nothing fancy—maybe a few hundred lines per month. The old pricing was straightforward per second. Now, it seems like the voices I was using are now classified as "Pro," with a significant jump in cost per second.
Has anyone else experienced this? I'm trying to figure out:
* Did all existing voices get moved to the Pro tier?
* Is there a clear list of which voices are Standard vs. Pro now?
* Are there any workarounds, like API parameters, to use a standard voice if one exists for my use case?
I love testing new features, but a pricing change this drastic without a very clear communication and migration path is a rough developer experience. I might have to pause my integration and look at alternatives, which is a shame because the quality was great for our needs.
Ship fast, measure faster.
Yeah, I saw the email about the new tiers but didn't think my usage would get hit. Sorry that happened.
I'm also using it for system alerts. Did you check if your specific voice ID is now listed under Pro in their docs? I had to dig a bit to find the updated voice list. It wasn't super clear.
Makes me wonder if there are any other services people are switching to for this kind of thing.
CloudNewbie
Totally feel you on the docs not being clear. I had the same hunt. The "Pro" voices list was tucked away in a blog post, not the main API reference.
I'm sticking with them for now because their latency is great for our alerts, but I've started slicing down usage. Stuff like caching repeated alerts and shortening script phrases.
For alternatives, I've heard folks mention ElevenLabs and Play.ht for similar use cases, but I haven't done a proper cost comparison yet. Have you tried any others?
null
Caching is a solid move. For our alert system, we started storing common phrases like "CPU threshold exceeded" as static audio files in S3 and serving them directly. The build pipeline generates them once, and it cut our API calls by about 70%.
> latency is great for our alerts
That's key, though. We tried Play.ht last year for a test project, and while the cost was better, the extra 200-300ms on generation was a dealbreaker for real-time alerts. ElevenLabs felt quicker, but we ran into some watermarking issues on their lower tiers. Have you benchmarked the actual latency of the cached vs. live approach with Resemble? Sometimes the fetch from storage can add its own delay.
Pipeline Pilot
Oh, the classic silent re-tiering. Happens every time a platform decides its initial "cost-effective" pricing was too generous. I'm not surprised in the slightest.
To your specific question about parameters or workarounds: no, there usually isn't. Once a vendor flips the switch on a voice ID's classification, your API calls are billed at the new rate, full stop. Your only real lever is to find a voice that's still "Standard" from their list and switch your integration, but the whole point is they want you on the pricier tier. That's the migration path: pay up.
Have you actually calculated the per-alert cost now versus before? I'd bet for "a few hundred lines," you're talking about a jump from pennies to dollars, which for system alerts is frankly absurd. You could host your own lightweight TTS container for that.
Your k8s cluster is 40% idle.