Ask
26

Voice generation bill got out of hand: the four options and what each one costs you

Narration went from a hobby line to my largest bill. The choice is less about price per character than which compromise you take.

  • Hosted premium. Still audibly ahead for anything longer than a couple of minutes. Priced per character, which is fine for interface prompts and brutal for narration, because the bill scales exactly with ambition.
  • Big cloud voices. Much cheaper and much better than their reputation. The compromise is prosody across a paragraph: sentences are good, something drifts over long passages. Nobody notices in an interface; some people notice immediately in narration. Test on your audience, not on yourself.
  • Open model locally. Free per character once you have the hardware and patience. Right answer if you produce a lot regularly, wrong one if you need twenty seconds occasionally.
  • Hire a person. For a short, fixed, high-value script heard thousands of times, often cheaper than all of the above and better than any of them.

Check before you switch anything: the commercial terms, before the price. Some of the cheapest have restrictions on how output may be used. That mistake is far more expensive than picking the wrong tier and it is one page of reading.

3 answers Share
Report

Answering anonymously, a moderator will review it first.

  • @licence_the_catch · 3w ago

    What actually varies: commercial use at all, use in something you charge for versus something with ads, attribution, whether audio made on an old plan survives a downgrade, and whether cloned or custom voices carry different terms from stock ones. That last one catches people, because the custom voice is what you built the project around.

    23
    Share
    Reply

    Answering anonymously, a moderator will review it first.

    Report
  • @local_voice_now · 3w ago

    Running locally, the real cost is consistency, not hardware. One paragraph is easy; forty minutes sounding like one person is the job.

    Fixes: generate in larger chunks than feels necessary, keep every parameter identical across a project including the reference clip, normalise afterwards so level changes are not doing the work of tone changes.

    17
    Share
    Reply

    Answering anonymously, a moderator will review it first.

    Report
  • @tts_bill_shock · 3w ago

    The thing that cut my bill most was not switching provider, it was caching on a hash of the text plus settings. Most of my regeneration turned out to be accidental. Measure that before migrating, or the habit follows you.

    1
    Share
    Reply

    Answering anonymously, a moderator will review it first.

    Report