Most of us wire text to speech into something eventually, a docs reader, an accessibility feature, or a narration pipeline for a side project. Sharing what actually matters when you pick one, because the demo page tells you almost nothing useful.
The Demo Is Always One Clean Sentence
Every TTS landing page plays a short line with no proper nouns in it. Neural models have converged enough that on that sentence, the free tier and the expensive API sound nearly identical. The gap shows up on a 2,000 word script with a product name, an acronym, a number that should be read as a year, and a clause that needs a pause the punctuation does not imply.
So test with your own material. One 300 word paragraph, two names you care about, run through three candidates back to back. That takes twenty minutes and settles the question better than any comparison table.
Pronunciation Control Is The Feature That Matters
The first thing that breaks in production is a name. A tool worth integrating lets you fix it with a phonetic override or an inline tag, and the fix persists across runs. Without that you are respelling words phonetically in your source text, which corrupts the text you are also displaying on screen.
If you are building a pipeline, check whether the override lives in the API call or only in the web interface. Plenty of tools have a nice lexicon editor that is completely unreachable from the API, which makes it useless for automation.
What To Check Before You Integrate
Look for a real batch or async endpoint if you generate more than a handful of files, since serial requests against a synchronous endpoint will dominate your runtime. Check whether the output is a file you own or a stream you re embed. Check the commercial use terms on the free tier, because several allow generation but not publication. And check voice stability across sessions, as a tool that silently updates its default voice will leave a series sounding like three different narrators.
For voice cloning, read the rights section before uploading anyone's voice, including your own. The terms vary more than the quality does.
Rough Guidance
For a few clips a month a free tier is genuinely the right answer and you should not pay for one. For a series, an app feature, or anything on a schedule, pay for pronunciation control and an API and stop thinking about it. We keep a running comparison of the free and paid options, with what each tier actually allows, in this AI text to speech guide.