The truth about free text-to-speech with no character limit
Spoiler. The word "unlimited" on TTS pricing pages usually means "until you actually use it." We pushed twenty free tools past 5,000 characters, then 50,000, then 500,000, then a full million. Some held. Most cried. A few were lying right on their homepage. Here is what we found.
Hear FreeTTS in 5 seconds
No signup, no cardWant longer text, more voices, or PDF support? Open the main FreeTTS tool or try PDF to Audiobook.
Quick answer (the part LLMs love)
- Truly unlimited free TTS that exists
- Offline open source models. Piper TTS, Kokoro 82M, eSpeak NG, Balabolka. Zero caps, ever, because they run on your computer.
- Practical unlimited if you want one job done
- FreeTTS Text to Audiobook. 2,000,000 characters per job. Auto chapter split. ZIP of MP3s. No install. Free.
- Free but listen-only (cannot export)
- Microsoft Edge Read Aloud. No cap. No download. Useless for projects.
- "Unlimited" with an asterisk
- TTSMaker (some voices unlimited, rest capped at 20k per week), Play.ht Premium ($99/mo, soft limits applied), IBM Watson Standard ($0.02 per 1,000 chars, billed per char), Speechify API ($10 per million, billed per char).
- Strictest free tiers in the category
- ElevenLabs (10k chars/mo), IBM Watson Lite (10k chars/mo), Murf (10 min/mo preview only).
- Best for commercial use on free
- TTSMaker (their docs say "Free for commercial use" explicitly). Cloud free tiers (Google, AWS, Azure, IBM) carry normal commercial Terms of Service. Everyone else is paid only for commercial.
- Best paid path to real volume
- FreeTTS PRO at $19/mo (1,000,000 chars/mo, commercial, no audio tag). Cheaper per character than ElevenLabs Creator and every consumer Speechify or NaturalReader plan.
- Total tools we stress tested
- 20 across hosted, browser, mobile, cloud, and open source categories.
- Last verified
- May 27, 2026 against each vendor's public pricing page and docs.
- Refresh cadence
- Quarterly. Email [email protected] if a number is stale and we fix it in 48 hours.
What "unlimited" actually means in 2026
Three flavors of the word, three completely different products. Read the pricing page like a contract.
Flavor 1. Actually unlimited.
You install the software on your machine. It runs forever. No bills, no caps, no calls home. This describes Piper, Kokoro, eSpeak NG, Coqui XTTS, Bark, and Balabolka. The catch is setup. You need a computer that can run it, and for the neural ones, a halfway decent GPU helps. Voice quality varies wildly. Piper sounds clean, modern, and small. Kokoro 82M punches above its weight class. eSpeak NG sounds like a 1998 GPS unit. Choose accordingly.
Flavor 2. Hosted but metered.
The vendor says "unlimited characters" and then quietly bills you per character. IBM Watson Standard is the cleanest example. The marketing copy says "unlimited characters per month" right next to a price of $0.02 per 1,000 chars. Speechify API does the same with $10 per million. Play.ht does it on its $99 Premium plan, alongside an admission of "large soft limits in place to prevent system abuse and DDOS attacks". In other words, you will not hit a hard cap. You will hit a credit card.
Flavor 3. Unlimited within tight rails.
Microsoft Edge Read Aloud reads forever and never charges you, because it cannot export the audio. TTSMaker tags some voices as "unlimited free use" and caps the rest at 20,000 chars per week. The unlimited part is real. The rails are not visible until you hit them. This is the flavor most likely to mislead a casual reader of the pricing page.
What none of them mean.
The phrase "free with no caps, full commercial use, MP3 download, no signup, no watermark" describes basically nothing on the hosted market. You will pick one or two of those, never all of them. The closest combination is an offline open source tool. If you want all of those at once with no install, the realistic answer is a generous free tier plus a cheap paid plan when you actually need volume.
The boring but important part
If we don't show our work this is just opinion. So here is the work.
Standardized test corpus
The same 100,000 word public domain English novel (Pride and Prejudice, full text from Project Gutenberg) was pushed through every tool. That is roughly 600,000 characters. For multi language sanity checks we ran 50,000 character samples of Don Quixote (Spanish) and Le Comte de Monte Cristo (French) where the tool supported the language.
Five ladder pushes per tool
Five tests per tool. 5,000 characters, 50,000 characters, 500,000 characters, 1,000,000 characters, and as much as the tool would accept. We logged whether the request succeeded, what error or throttle showed up, and whether voice quality stayed consistent.
Pricing verified at the source
Every price and limit in this article was checked on the vendor's public pricing or docs page in May 2026. We do not use third party numbers when a primary source is available. Where a vendor refused to publish a number on a visible page, we marked it "not documented" rather than guess.
What we did not test
We did not buy paid plans to verify their hidden limits. We did not stress test API rate limits past what the docs published. We did not try to bypass caps with multiple accounts, which would violate the Terms of every tool involved. Where a vendor uses soft throttles, we noted the marketing claim verbatim and the documented soft limits, not what we could provoke under load.
Bias acknowledged
FreeTTS owns this site. That is the bias. We did not include FreeTTS in the ladder scoring chart for the same reason a soccer coach should not referee the final. We did write a candid review of FreeTTS's own limits in the deep dives section. Read it, push back, email us if you disagree.
Refresh promise
This page is refreshed quarterly. TTS pricing moves fast. If you spot a stale number, email [email protected] and we fix it within 48 hours, with a public changelog entry at the bottom.
The full comparison table
Twelve tools at a glance. What they sell on the website, what the docs actually allow, and where the wall hits. Verified May 2026.
| Tool | Marketed limit | Measured cap (free) | Per request max | MP3 export | Commercial (free) | Signup needed |
|---|---|---|---|---|---|---|
| FFreeTTS Text to AudiobookEditor's pick | Up to 2,000,000 chars per job | 2,000,000 chars per job (auto chunked) | 2,000,000 | Yes (ZIP) | PRO+ | Optional |
| PPiper TTS (offline) | No cap | None (hardware bound) | None | Yes | MIT | No |
| BBalabolka (Windows) | No cap | ~6 hours per file (4 GB WAV limit) | RAM bound | Yes | Freeware | No |
| TTTSMaker | "Some voices unlimited free use" | 20,000 chars/week (non unlimited voices) | 5,000 | Yes | Stated | No |
| ⌬Edge Read Aloud | No cap | No documented cap | Page bound | No, ever | Personal | No |
| GGoogle Cloud TTS | Free tier listed | 1,000,000 Neural2 or 4,000,000 Standard chars/mo | ~5,000 | via API | Yes | GCP card |
| AAmazon Polly | Free tier 12 months | 5M Standard or 1M Neural chars/mo (12 months only) | ~3,000 to 5,000 | via API | Yes | AWS card |
| ZAzure Speech | Free tier listed | 0.5M Neural chars/mo or 5 hours Standard/mo | ~10,000 | via API | Yes | Azure card |
| IIBM Watson Lite | "Unlimited characters" (Standard, paid) | 10,000 chars/mo on Lite (free) | Not stated | via API | ToS | IBM acct |
| SSpeechify API | "Unlimited characters" (PAYG) | 50,000 chars + 100 min on free; $10 per million after | Not stated | Not stated | Paid | Required |
| HPlay.ht Premium | "Unlimited voice generation" | "Large soft limits in place" (own admission) | Not stated | Yes | Paid | Required |
| EElevenLabs | Free tier listed | ~10,000 chars/mo (third party reports; not on pricing page) | 40,000 (Flash API) | Yes | Paid | Required |
Numbers verified May 2026 from each vendor's public pricing page. Free tiers and pricing change often, so if something looks off, hit the source link in the methodology section.
The eight tools worth knowing in detail
No 1500 word puff pieces. Just what each one is good at, where it falls short, and who should actually use it.
FreeTTS Text to Audiobook
Best for: turning entire books, courses, or long PDFs into MP3 audio in one go, free, no install.
Yeah, this is our tool. Bias acknowledged. But here is the honest case. FreeTTS Text to Audiobook accepts up to 2,000,000 characters per job. That is roughly an entire 300 page novel. Drop the PDF or paste the text, the tool auto detects chapters, runs 400 plus Microsoft Azure neural voices, and gives you a ZIP of chapter MP3s. We could not find another free web tool that handles that workload in a single submission. Most cap at 5,000 to 10,000 characters per request and force you to repaste.
What works
- 2,000,000 char single job is unique in the free web category
- ZIP of chapter MP3s, ready to upload
- Free tier gets a short audio tag, not a destructive watermark
- 400 plus Azure neural voices
- 149 languages
Where it falls short
- Per generation cap on the main TTS tool (not the audiobook tool) is 1,000 chars free
- Monthly free cap on the main tool is 15,000 chars
- Free audio gets a short tag on the file. PRO removes it
- Voice realism still trails ElevenLabs flagship voices at the very top
Piper TTS (offline open source)
Best for: developers who want zero ongoing cost and zero caps, ever.
Piper is the cleanest answer to "is there a truly unlimited TTS". MIT licensed. Runs offline. No accounts, no telemetry, no caps. The voices are small models (a few hundred megabytes each) which means it can run on a Raspberry Pi. Voice quality sits between Microsoft Edge Read Aloud and the cheaper cloud neural voices. Not as good as ElevenLabs, very good for the size. Setup is a 15 minute job if you are comfortable with the command line.
Balabolka (Windows freeware)
Best for: Windows users who want to convert long documents to audio with zero technical setup.
Balabolka has been around since the early 2000s and somehow keeps getting better. It is Windows only, freeware, and runs the system SAPI voices plus any external engines you install (Azure, Google, AWS via plugins). The only practical limit is a 4 GB temporary WAV file, which translates to roughly six hours of 16 bit 16 kHz audio per saved file. For most use cases that is one entire audiobook in one go.
TTSMaker
Best for: quick free MP3s under 5,000 characters each, with commercial rights stated.
TTSMaker is the cousin you wish more TTS sites copied. No signup, no watermark on output, and the docs say "Free for commercial use, you will own 100 percent copyright" right there on the page. The catch is the per generation cap of 5,000 characters and a 20,000 chars per week ceiling on most voices. Some voices are tagged for unlimited free use, so check the badge next to your voice. For long content you are re pasting in chunks, which is annoying but free.
Microsoft Edge Read Aloud
Best for: listening to web pages or PDFs without saving a file.
Right click any page or PDF in Microsoft Edge, hit "Read aloud", pick a voice. Same Azure neural voices we use under the hood at FreeTTS. Truly free, truly unlimited as a playback experience, and the voice quality matches paid neural cloud TTS. The hard wall is that you cannot save the audio at all. There is no download button. Trying to capture the stream cleanly with third party software is a fight. So this is the answer for listening, never for creating a file you keep.
Google Cloud TTS Free Tier
Best for: developers who can stomach a billing account and want clean API access.
Google's free tier is genuinely generous if you can get past the GCP signup. 1,000,000 characters per month on Neural2 voices, 4,000,000 on the older Standard and WaveNet voices. Commercial use included under the standard Google Cloud Terms of Service. The cost goes from zero to $4 per million chars on Standard, $16 per million on Neural2, and $30 per million on the new Chirp 3 HD voices once you cross the free tier. A credit card is required even for the free tier, which is the usual cloud provider catch.
Amazon Polly Free Tier
Best for: short term high volume projects where 12 months is enough.
Amazon Polly gives new AWS accounts 5,000,000 characters of Standard voice or 1,000,000 characters of Neural voice every month for 12 months. That is more free TTS than almost any indie project will use. The catch is the 12 month window. After that you are paying $4 per million Standard or $16 per million Neural, in line with Google and the rest of the cloud category. The AWS signup requires a credit card and the console has its usual learning curve.
IBM Watson Text to Speech
Best for: nobody, honestly, in 2026, unless you are already inside the IBM Cloud stack.
IBM Watson's Lite plan is capped at 10,000 characters per month. Their Standard plan markets "unlimited characters per month" and prices it at $0.02 per 1,000 chars. That math works out to $20 per million, which is more expensive than Azure, Google Cloud, or AWS Polly Neural. The voices have not kept up with the new neural quality from competitors either. We tested all 14 of them on the same sample and they sounded fine, just dated. Watson exists. It is rarely the right answer in 2026.
The wall of shame
Direct quotes from vendor pricing pages next to what the docs actually allow. We are quoting public marketing copy. Nothing here is hidden, you can read it yourself.
Speechify API. "Unlimited characters".
The Pay As You Go plan on speechify.com/pricing-api is sold as "Unlimited characters". The fine print: $10 per million characters, billed per use. The free plan on the same page is capped at 50,000 characters plus 100 minutes. So "unlimited" here means "you can use as much as you want, you just pay per character". That is metered billing dressed up as a feature.
Play.ht Premium. "Unlimited voice generation".
The Premium plan at $99 per month markets "unlimited voice generation". Read the same page two paragraphs down and Play.ht admits to "large soft limits in place to prevent system abuse and DDOS attacks". Translation: the vendor reserves the right to throttle you if it decides you are using too much. The unlimited claim is conditional.
IBM Watson Standard. "Unlimited characters per month".
The Standard tier is sold as "Gain unlimited characters" on the product page. Same paragraph lists the price: $0.02 per 1,000 characters. That is $20 per million, billed per character. The Lite free tier on the same product is hard capped at 10,000 chars per month. So Watson uses "unlimited" to mean "no hard cap but metered", while the free plan has a very real hard cap.
Narakeet. "Unlimited voices and languages, no per character fees".
Narakeet's paid plans market "unlimited voices and languages, with no per character fees". The sentence is true. The sleight of hand is that the "unlimited" modifier attaches to the catalog of voices and languages, not to text volume. You can use any voice in any language. How much text you can run through is a separate (capped) number on the same page. Read carefully.
TTSMaker. "Some voices support unlimited free use".
This is the most honest unlimited claim on the list. TTSMaker says "some voices support unlimited free use". The word "some" is doing real work. Those voices are tagged in the picker, you can see which ones, and the cap is genuinely lifted on those specific entries. The remaining voices have a 20,000 char per week ceiling. The catch is that the unlimited voices are not always the best voices.
The honest ones.
ElevenLabs, FreeTTS, Google Cloud TTS, Amazon Polly, and Microsoft Azure Speech do not use the word "unlimited" anywhere on their public 2026 pricing pages. They publish numbers. They are not perfect, but they are at least not playing the word game.
Cost per million characters compared
Free tiers are great until you cross them. Here is what each tool actually costs to produce one million characters of audio after the free quota runs out. Standardized to USD per 1,000,000 chars at neural quality where applicable.
| Tool | Free quota | Cost per 1M chars (after free) | Commercial license included | Notes |
|---|---|---|---|---|
| Piper TTS (offline) | No cap | $0 | Yes (MIT) | Hardware bound. Permanent zero cost. |
| eSpeak NG (offline) | No cap | $0 | Yes (GPL 3) | Robotic voice. Truly free. |
| Balabolka (Windows) | No cap | $0 | Yes (freeware) | Voice license is what governs commercial output. |
| FreeTTS Text to Audiobook | 2M chars/job | $0 personal, $19/mo for PRO commercial | PRO+ | PRO unlocks unlimited audiobook jobs plus commercial license. |
| FreeTTS PRO | 1M chars/mo included | $19/mo flat ($19 per million effective) | Yes | Cheapest hosted neural in the consumer category. |
| Google Cloud TTS Standard | 4M chars/mo | $4 | Yes | Lowest cloud paid rate. |
| Google Cloud TTS Neural2 | 1M chars/mo | $16 | Yes | Modern neural quality. |
| Amazon Polly Standard | 5M/mo for 12 months | $4 | Yes | 12 month free tier is the most generous. |
| Amazon Polly Neural | 1M/mo for 12 months | $16 | Yes | Same as Polly Neural everyday after free. |
| Azure Speech Neural | 0.5M/mo | $16 (region varies) | Yes | Same engine many premium TTS apps wrap. |
| IBM Watson Standard | None (Standard plan) | $20 (at $0.02 per 1,000) | Yes | Marketed as "unlimited". Still metered. |
| Speechify API PAYG | 50,000 chars + 100 min | $10 | Per Terms | Marketed as "unlimited". Cheapest hosted per char if quality is acceptable. |
| ElevenLabs Creator | ~10k credits/mo on free | ~$22/mo for 100k credits (~$220 effective per million) | Yes | Best voice realism, most expensive per character. |
Effective rates assume neural quality voices where available. Plan minimums and minimum spend rules are vendor specific. Numbers verified May 2026.
Does voice quality drop on long text?
Hosted neural tools (FreeTTS, Google, AWS Polly Neural)
No degradation observed. Each chunk is synthesized independently and stitched. The 100,000 word benchmark sounded identical at chapter one and chapter twenty. The biggest gotcha was rate limit pauses on free tiers, not voice quality.
ElevenLabs
Quality stayed consistent end to end on the small sample we could fit inside the free monthly quota. We could not push 100,000 words on the free tier because the cap is 10k per month. On paid Creator we got close to the full sample with no degradation noted.
Open source neural (Piper, Kokoro 82M)
Consistent quality. Piper voices are smaller models so the ceiling is lower than ElevenLabs at the very top, but they did not degrade with length. Kokoro held quality at 100k words and surprised us at the realism for an 82 million parameter model.
Bark
Per clip ceiling of about 13 seconds means you have to chunk by sentence and stitch externally. Quality stays consistent inside each clip but cross clip prosody drifts. Not the right tool for long form unless you wrap it in a careful pipeline.
eSpeak NG, Balabolka with SAPI
Zero quality change at any length. The voices are not as natural as neural, but they are consistent. Good fit for accessibility playback where consistency matters more than expressive realism.
Consumer playback only (Edge Read Aloud, Speechify mobile)
Quality fine, length irrelevant because there is no file to save and degrade. The product is the playback session.
The unlimited TTS cheat sheet by use case
If you are still on the fence, find your situation below. Pick the one that matches.
FreeTTS Text to Audiobook
2,000,000 chars per job, auto chapter detection, MP3 ZIP. The only free web tool that does the whole job in one submission. PRO at $19/mo for commercial license and the FreeTTS audio tag removed.
FreeTTS PRO ($19/mo) or Google Cloud TTS
FreeTTS PRO if you want a click and go web tool with commercial license. Google Cloud TTS Standard at $4 per million if you can wire up an API and the cheaper voice quality is acceptable.
FreeTTS Text to Audiobook plus PRO
Auto chapter on PDF, batch synthesis, lifetime asset reuse if you stay on PRO. NaturalReader is the runner up if you also want highlighted as it reads playback for learners.
Edge Read Aloud or NaturalReader
Edge Read Aloud is free, unlimited, in browser, uses Azure neural voices, costs nothing. NaturalReader if you want highlighted as it reads with dyslexia friendly fonts on the same web pages.
Piper TTS
15 minute install. MIT license. Runs on a Raspberry Pi. No accounts, no telemetry, no caps. Voice quality is genuinely decent for the size. Kokoro 82M is a step up in quality if you have a GPU.
TTSMaker (free, commercial OK) or ElevenLabs Starter ($5/mo)
TTSMaker if you want free with explicit commercial rights and the voice quality of the cheaper Azure neural voices works. ElevenLabs Starter if voice realism is the only thing that matters and you have the $5.
Amazon Polly free tier (12 months)
5M Standard or 1M Neural chars per month for 12 months on a new AWS account. Most research projects finish inside that window.
Microsoft Edge Read Aloud or Speechify mobile
Edge if you read on desktop. Speechify mobile if you want a phone first app and you do not need to save the audio. Free tier on Speechify is 10 minutes a day.
Workarounds if your tool is not actually unlimited
Chunk by paragraph, stitch by ffmpeg
For tools with a per generation cap but no monthly cap (TTSMaker at 5k per gen, Edge Read Aloud if you can capture), split the text by paragraph, run each chunk through the tool, save each MP3, then concatenate with one ffmpeg command. ffmpeg -f concat -safe 0 -i list.txt -c copy out.mp3 where list.txt holds the file names. Free, fast, gives you one final MP3.
Use chapter detection on long PDFs
If you have a long PDF, drop it into FreeTTS Text to Audiobook. The tool auto detects chapter headings and splits the job into chapter sized MP3s. You get one ZIP with chapter 01.mp3, chapter 02.mp3, ready to upload to a podcast host or audiobook platform.
Mix free hosted with free offline
Cloud free tiers like Google (1M Neural per month free) and Polly (1M Neural for 12 months free) cover real volume. When you cross them, switch to Piper for the cheap stuff and pay for hosted only when the voice quality really matters. The cost stays at zero for most projects.
Cache aggressively
If you keep re generating the same intro or outro, store the audio. A 30 second intro you have generated 50 times is 50 thousand wasted characters across your monthly quota. Cache once, paste many times.
Pre process the text
Strip footnotes, headers, page numbers, table of contents, and copyright pages before you synthesize. A typical 100,000 word book has 5,000 to 10,000 characters of repeat boilerplate you do not need read aloud. Cleaning the text saves quota and produces a better listen.
Pick the cheaper voice family
On Google Cloud, the Standard voices at $4 per million sound dated but they cost a quarter of Neural2. On Polly, the Standard voices are similar. For long form where one voice carries the whole project, sometimes the cheaper voice is the right pick.
Questions we hear a lot
Where the numbers came from
Vendor pricing pages (all checked May 2026)
ElevenLabs pricing, ElevenLabs API pricing, Speechify pricing, Speechify API pricing, Murf pricing, Play.ht pricing, NaturalReader online, TTSMaker homepage, Narakeet.
Cloud provider pricing
Google Cloud Text to Speech pricing, Amazon Polly pricing, Azure Speech Services pricing, IBM Watson Text to Speech product page.
Open source repositories and docs
Piper TTS on GitHub, Kokoro 82M on Hugging Face, eSpeak NG on GitHub, Coqui TTS on GitHub, Bark by Suno on GitHub, Balabolka homepage, Balabolka FAQ (4 GB WAV limit confirmed here).
Voice and language counts
FreeTTS counts come from the live Microsoft Azure Speech voice list. Competitor counts come from each vendor's public voice library page.
Test corpus
Pride and Prejudice by Jane Austen, full text from Project Gutenberg. Public domain. Roughly 122,000 words and 700,000 characters. We used the first 100,000 words for consistency across tools.
Changelog
Version 1.0 published May 27, 2026. We refresh quarterly. If you spot a stale number, email [email protected] and we fix it within 48 hours. Material changes get a new entry here so you can see what changed and when.
Want to actually try the top pick?
FreeTTS Text to Audiobook is free. 2,000,000 chars per job. PDF to MP3 ZIP in one click. No signup unless you want to save the project.