Free AI voice credits: what one finished narration actually costs

By Rita

AI Toolbox

A free voice plan does not ration minutes. It rations characters and credits — and one finished minute of narration usually takes several generations. So the question is never “how many minutes do I get?” but “how many characters does one finished minute cost me, and how many credits does each character cost?” Answer those two numbers and you can tell in five minutes whether a free plan covers your week.

Count minutes per finished narration, not narrations per month

Every click on Generate is a spend, but only the keeper counts as output. A two-minute narration is not one generation. It is a draft, then a fix for a mispronounced name, then a retake for pacing, then a fix for a sentence that ran too fast — four or five spends for one finished minute. Budgeting by “minutes per month” hides that multiplier; budgeting by generations per finished minute exposes it.

The arithmetic is simple, and it has only two inputs:

Finished minutes per allowance = monthly or weekly characters ÷ (characters per minute × generations per finished minute).

Characters per minute comes from the tool: spoken English runs roughly 700 to 1,000 characters a minute depending on voice and speed, and TTSMaker's own FAQ puts one million characters at about 23 hours on average. Generations per finished minute comes from you: it is whatever your last real job actually took. Measure the second number once, and the first number stops being a mystery.

What the official pages say — checked 2026-11-06

Four free plans, four different shapes of limit. The shape matters more than the headline allowance.

ElevenLabs Free: the countable one — 10,000 credits a month

ElevenLabs' own pricing page gives free accounts 10,000 credits a month, and text to speech costs about 1 credit per character, so the allowance is roughly ten minutes of narration a month. The credits are one shared pool across every product: the same 10,000 pays for text to speech, speech to text, music, sound effects and dubbing, so a side experiment leaves fewer credits for narration. Speech to text costs 330 credits a minute, music 900 a minute, sound effects 200 per generation, and dubbing 2,000 to 10,000 a minute depending on the mode. The free plan includes 3 Studio projects. Two hard limits: the credits do not roll over, and the free plan has no commercial license — the commercial license starts on the paid Starter plan. Official pricing page.

This is the only free plan of the four where you can do the division before you start: 10,000 credits, divided by the cost the plan page states, divided by the generations your job takes.

TTSMaker Free: the countable weekly one — 20,000 characters a week

TTSMaker's own pricing FAQ gives its free plan 20,000 characters a week, with no sign-up and no card required, and each free conversion is capped at 500 characters, so a long script has to be split into short passages. Voices marked Unlimited do not count against the quota. By TTSMaker's own conversion math — 300,000 characters is about 6.9 hours of English audio — the weekly allowance is roughly 25 to 30 minutes of English narration a week. Downloads are free, and its FAQ says its voices can be used in YouTube and other videos and points to its copyright and commercial license terms. Official pricing FAQ.

The weekly reset is the useful part: a missed week is a missed allowance, but a heavy week is never more than seven days away from a refill.

Google Cloud Text-to-Speech: the biggest allowance, behind a billing account

Google's own pricing page gives Standard and WaveNet voices up to 4 million free characters a month, and Neural2, Chirp 3 HD and Studio voices up to 1 million each. The newest Gemini-TTS models have no free allowance. Two catches the page states plainly: you must enable billing to use the service at all, and usage past the free allowance is charged automatically. Characters are counted as submitted, including spaces, newlines and SSML tags. Official pricing page.

The allowance is enormous next to the other three, but it is a free tier inside a paid cloud account, not a no-card free plan. If you are not already a Google Cloud user, the setup cost is an account, a project and a payment method — and an overrun bills itself.

Azure AI Speech F0: the developer-shaped one — 500,000 characters a month

Microsoft's pricing page lists 500,000 neural text-to-speech characters a month free on the F0 tier, plus 5 audio hours a month of real-time speech to text. It is an API service, not a web editor: you create a Speech resource in the Azure portal and call it from code or an app, and an Azure subscription is needed even for the free tier. Official pricing page.

Half a million characters is about eight to twelve hours of narration a month — far more than any web editor's free plan. The price is setup: this tier fits a developer wiring narration into an app, not someone pasting a script into a web page.

Measure it on your first real job

Do not estimate generations per finished minute. Measure it, once, on a job you actually need:

1. Write the full script before the first generation — the exact words, the names spelled out, and the pauses marked.
2. Generate, then tally every regeneration until you have a keeper: mispronunciations, pacing fixes, retakes for tone, all of them.
3. That tally is your generations per finished minute. For example, a two-minute video intro might take six: one draft, two pronunciation fixes, two pacing retakes, one tone change.
4. Multiply by the characters per generation, and compare with the weekly or monthly allowance.

If the result does not fit the allowance, that is a tool-fit signal, not a scripting failure. Either the job needs a tool with a countable allowance, or it needs fewer generations — which is a workflow fix, covered below. Do not solve a credit shortage by re-rolling harder; re-rolling is what caused it.

Four ways to spend fewer credits

1. Write the whole script before the first run. Every “make it sound warmer” re-roll is a spend. Decide the words, the names and the pauses first, then generate once.
2. Fix pronunciation without re-rolling the whole take. Most generators let you respell a name phonetically or mark it with SSML instead of regenerating the full passage. Re-rolling a two-minute take to fix one name burns the whole take; a phonetic spelling burns one short test.
3. Change one thing per regeneration. Voice, speed, or wording — not all three. One change per run teaches you what worked; three changes per run teaches you nothing and costs three times as much.
4. Batch similar jobs in one session. While the voice, speed and script style are fresh in your head, the second and third narrations in a set take fewer generations than the first. Spreading one set across a week re-pays the learning cost every time.

When free stops being enough

Two signs, and only two, mean the free tier is the wrong tier rather than a tight one. First, the same real task fails twice in a week for credits, not for quality — the allowance, not your scripting, is the bottleneck. Second, the free tier's other conditions block the job: ElevenLabs' free plan has no commercial license, so monetized video does not belong there, and Google Cloud's free tier needs a billing account, so a no-card workflow does not belong there either. Our commercial-use check covers the license question in full. A paid plan fixes the first problem; it is the only fix for the second that also changes the terms.

Reviewed 2026-11-06. Free tiers and trial terms change; the provider's own plan page and the cost shown before you generate are authoritative.

← AI Toolbox · How we verify offers