The best ElevenLabs alternative depends on the job. For narration and voiceovers, look at Murf, WellSaid Labs or Speechify Studio. To dub videos into other languages, HeyGen or Rask AI. For an API behind an app or a voice agent, Cartesia, Google Cloud Text-to-Speech, Azure AI Speech, Amazon Polly or OpenAI. To listen to documents, NaturalReader or the Speechify app. To run voices on your own machine, Chatterbox, Kokoro or Piper. Below: the documented limits behind each choice, a comparison table and a plan for moving your projects.
Why people look for an ElevenLabs alternative
ElevenLabs is where many people start with AI voices, and its own pages explain most of the reasons they compare it with other tools. As of November 2, 2026:
- The free plan has 10,000 credits a month, shared by every product: text to speech, dubbing, music, sound effects and the rest. Unused credits don't roll over on the free plan.
- Free-plan audio isn't for commercial use. If you publish it, ElevenLabs asks you to credit it with "elevenlabs.io" or "11.ai" in the title. Content made outside a paid subscription can't be used commercially later, even after you upgrade.
- The commercial license and Instant Voice Cloning start with the Starter plan; Professional Voice Cloning starts with Creator.
- Data is stored in the US as standard. Isolated environments in the EU, India and Singapore exist, but only for Enterprise customers, and they are separate accounts with their own API keys.
- Voice clones can't be exported. They stay in your ElevenLabs account.
None of this makes ElevenLabs a poor choice. It means the right alternative depends on what you need: commercial rights from day one, data that stays in Europe, or an open model you run yourself. If the terms are new to you, our guide to what TTS is explains how text-to-speech works.
The short answer by situation
| Your situation | Look at first | Why |
|---|---|---|
| Voiceovers for videos, courses and ads | Murf, WellSaid Labs, Speechify Studio | Studio editors with commercial rights on paid plans |
| Clone your own voice | Speechify API, HeyGen, Cartesia | Self-serve cloning; Speechify and HeyGen describe their consent checks |
| Dub a video into other languages | HeyGen, Rask AI | Translation, voice cloning and lip sync in one tool |
| API for an app or a voice agent | Cartesia, Google Cloud TTS, Azure AI Speech, Amazon Polly, OpenAI | Developer APIs; all but OpenAI have a free plan, tier or credits |
| Data must stay in Europe | Google Cloud TTS, Azure AI Speech | You pick a European endpoint or region yourself; elsewhere it's an Enterprise option |
| Listen to PDFs, e-books and web pages | NaturalReader, Speechify app | Reading apps with free plans for personal listening |
| Run voices offline or on your own server | Chatterbox, Kokoro, Piper | Open licenses; each model and voice has its own terms |
How we chose these alternatives
We started from the jobs people use ElevenLabs for. For each tool we read the vendor's own pricing, help and documentation pages on November 2, 2026, and noted six points: the free plan or trial, German, French and Spanish support, cloning and its consent checks, commercial use, the API and where data is stored.
We left out tools whose website didn't load and mark those that no longer take new customers. The list is grouped by job, not ranked: how a voice sounds depends on your script and language, so listen to each candidate with your own text. Prices change often and aren't listed; plan names come from the vendors' pages.
Narration and voiceover: Murf, WellSaid Labs, Speechify Studio
These three are closest to ElevenLabs' studio: paste a script, pick a voice, adjust pacing, export.
Murf
- Free option: the Studio free trial includes 10 minutes of voice generation and a shareable preview. Downloading and exporting need a paid plan.
- Languages: Murf lists 35+ languages on its text-to-speech page, German, French and Spanish among them.
- Commercial use: Murf says the commercial rights for voiceovers you create stay valid after your subscription ends.
- Cloning: only for Enterprise customers, through the API. Instant clones use one sample of up to 30 seconds. Murf asks you to clone only a voice you own or have clear, documented consent to clone.
- API and data: the API has a free trial credit without a credit card. For its Falcon 2.1 streaming voices, Murf documents 11 regional environments, including EU (Central) and the UK, listed under Enterprise; account data such as billing stays in the US.
WellSaid Labs
- Free option: 3 download minutes a month, without commercial rights.
- Languages: the Starter and Pro plans include all English voices. Spanish, French, German and other languages are part of the Enterprise plan.
- Commercial use: full commercial rights on paid plans.
- Cloning: WellSaid says users have never been able to clone voices on their own. Its voices come from actors who gave written consent and are paid for usage.
- API: you can request an API trial key after creating an account.
WellSaid fits English corporate narration; for other languages, ask about Enterprise.
Speechify Studio
- Free option: 600 Studio credits with the voiceover studio, dubbing studio and voice changer, but no commercial usage rights. Downloading your work as MP3 needs a paid plan.
- Languages: Speechify's voiceover FAQ lists English, Spanish, French, German and more.
- Commercial use: paid Studio plans include commercial rights in perpetuity for your own projects.
Voice cloning, and the consent checks behind it
Cloning is where the rules are strictest. Use your own voice, or that of someone who has given you written permission, and read the vendor's usage policy first. Whether a clone is lawful where you live is a legal question this guide can't answer, so check the vendor's terms as well.
| Tool | What you provide | Who can use it |
|---|---|---|
| Speechify API | The speaker reads a phrase Speechify issues; the recording (5 to 30 seconds) must match the voice sample and is kept as evidence | Paid API plans; not available in some jurisdictions |
| OpenAI custom voices | A consent recording with a fixed phrase (German, English, French, Spanish and more) plus a sample of 60 seconds or less | Eligible customers, through sales |
| Google Instant Custom Voice | A consent statement of up to 10 seconds plus up to 10 seconds of reference audio | Allow-listed users, through sales |
| Azure personal voice | A recorded consent statement plus a speech sample of 5 to 90 seconds | Approved use cases; you apply for access |
| HeyGen | HeyGen verifies voice ownership and requires written consent for third-party voices | One voice clone on the free plan |
| Murf | Clone only a voice you own or have documented consent for | Enterprise, API only |
| Cartesia | A clip of 10 to 60 seconds, recorded or uploaded; check its terms for consent rules | Pro plan and higher |
| ElevenLabs, for comparison | Professional clones only of your own voice, with verification | Starter (instant) and Creator (professional) |
Speechify switched off its older flow, where developers only asserted consent in a form field, on September 23, 2026. Now a clone is created only if the recording says exactly the issued phrase and comes from the same person as the sample; see Speechify's consent guide.
The open models Chatterbox and XTTS-v2 clone a voice from a short clip on your own computer. Nothing checks consent there, so the responsibility is yours; Chatterbox adds an inaudible watermark to every file.
Dubbing and video translation: HeyGen and Rask AI
These two cover ElevenLabs' dubbing job with video editing built in. For the difference between a voice-over and a full dub, see our guide to what dubbing is.
HeyGen
- Free option: 3 videos a month, watermarked, each up to 1 minute, in 30+ languages, with one voice clone.
- Paid plans: the Creator plan adds 175+ languages and dialects and removes HeyGen branding.
- Translation: upload a video or paste a YouTube link and choose up to 10 target languages at once. HeyGen keeps the speaker's voice and syncs the lips; editing and proofreading the translated script is listed as a Pro plan feature.
- Consent: see the cloning table above.
Rask AI
- Free option: a free trial with 3 minutes included and no credit card.
- Languages: dubbing in 135+ languages, voice cloning in 57.
- Plans: lip sync starts with the Creator Pro plan. API access and custom integrations are part of Enterprise.
With either tool, have someone who speaks the target language check the translated script before you publish; in German, French and Spanish, formal or informal address changes the whole tone.
Voice agents and developer APIs
If ElevenLabs runs inside your product, look at these five developer APIs.
Cartesia
- Free option: 20,000 credits a month. The commercial use license and instant cloning start with the Pro plan; professional cloning starts with Startup.
- Languages: Sonic 3.6, released August 27, 2026, speaks 44 languages, including German, French and Spanish, and accepts regional locale codes.
- Data: dedicated regional deployments in the EU, UK, US, India and Australia are available to Enterprise customers.
OpenAI
- Models:
tts-1,tts-1-hdandgpt-4o-mini-ttsare deprecated and shut down on January 6, 2027. OpenAI points new work togpt-realtime-2.1-miniin the Realtime API, a different workflow. - Rules: OpenAI's usage policies require you to tell listeners clearly that the voice is AI-generated.
- Languages: support follows the Whisper model, which includes German, French and Spanish; OpenAI says the voices are optimized for English.
- Data: data residency, including a Europe region (EEA plus Switzerland), for eligible customers.
Google Cloud Text-to-Speech
- Free option: you must enable billing, then each month the first 1 million characters of Chirp 3: HD voices and the first 4 million of Standard and WaveNet voices are free. Instant custom voices have no free usage.
- Languages: Chirp 3: HD voices cover German (de-DE), French (fr-FR, fr-CA) and Spanish (es-ES, es-US). Google also offers Gemini-TTS models in the same service.
- Data: with the
euendpoint, Google says data at rest and in use stays within Europe; regional endpoints include London, Frankfurt and Eemshaven.
Azure AI Speech
- Free option: the F0 tier includes 0.5 million characters of neural text to speech a month.
- Languages: voices for de-DE, fr-FR, es-ES and es-MX, among many others.
- Data: you create your Speech resource in a region, and the list includes West Europe, North Europe, France Central, Germany West Central and Switzerland North.
- Cloning: personal voice and professional custom voices are limited-access features.
Amazon Polly
- Free option: the free tier includes 5 million characters a month for Standard voices and 1 million a month for Neural voices during the first 12 months. AWS accounts created since July 15, 2025 receive Free Tier credits that can be spent on Polly.
- Languages: German (Germany, Austria, Switzerland), French (France, Belgium, Canada) and Spanish (Spain, Mexico).
- Pronunciation: custom lexicons follow the W3C Pronunciation Lexicon Specification (PLS), the same format ElevenLabs exports, and are stored per AWS region.
- Custom voices: Brand Voice is a custom project with the Polly team, not a self-serve feature.
Reading documents aloud: NaturalReader and Speechify
If you mainly listen to PDFs and articles, a reading app fits better than a voice studio.
NaturalReader has a free plan with no time limit and lists 90+ languages. Its personal version reads PDF, EPUB without DRM, Word, PowerPoint, web pages and more; scanned pages need OCR, which comes with a subscription. Its audio is for your own listening only: sharing MP3 files, posting audio online or using it in videos needs the separate commercial product.
Speechify's reading app has a free plan with 10 voices and speeds up to 1.5x. Premium adds 1,000+ voices, 60+ languages and Scan & Listen for printed pages. Our guide to Speechify alternatives compares more readers for every device.
Before paying for any reader, check what your device already does: our guide to reading text aloud for free covers the built-in options on Windows, Mac, phones and browsers.
Open-source and self-hosted voices
Open models run on your own computer or server, so your scripts never leave it and there's no account or credit limit. In exchange, you handle setup, hardware and updates, and you must read the model's license and, for some projects, each voice's.
| Model | License | German, French, Spanish | Cloning | Worth knowing |
|---|---|---|---|---|
| Chatterbox (Resemble AI) | MIT | All three among 23+ languages in Multilingual V3; packs for Spain and Latin American Spanish | Yes, from a reference clip | Every file carries an inaudible watermark |
| Kokoro-82M | Apache 2.0 | French and Spanish, no German | No, built-in voices only | Version 1.0 has 54 voices in 8 languages |
| Piper | GPL-3.0 | Voices for Germany, France, Spain, Mexico and Argentina | No, but you can train new voices | Each voice's model card states its own license |
| XTTS-v2 (Coqui) | Coqui Public Model License | All three among 17 languages | Yes, from a 6-second clip | Non-commercial use only |
A few details matter before you pick one:
- Chatterbox comes from Resemble AI. Next to the multilingual model, Chatterbox-Turbo is an English model built for low-latency voice agents.
- Kokoro is small, with 82 million parameters, and its Apache license allows commercial deployment, but it has no German voices.
- Piper now lives in a GPL-3.0 repository; the old one is archived. The project says it is intended for personal use and research, and some voices have restrictive licenses, so check each voice's model card.
- XTTS-v2 clones a voice from a clip of just six seconds, so the consent rule above applies in full. Its license allows only non-commercial use of the model and its output. The original Coqui repository is no longer maintained; a fork by the Idiap institute continues the code.
ElevenLabs alternatives compared
| Tool | Free option | German, French, Spanish | Cloning | Commercial use | API | Data in Europe |
|---|---|---|---|---|---|---|
| ElevenLabs (reference) | 10,000 credits a month | Yes | From Starter | Paid plans | Yes | EU environment, Enterprise |
| Murf | 10 minutes, no downloads | Yes | Enterprise, API | Paid plans | Yes | EU (Central) region, Enterprise |
| WellSaid Labs | 3 download minutes a month | Enterprise | No self-serve cloning | Paid plans | Trial key on request | Not stated |
| Speechify Studio | 600 credits | Yes | Not listed on Studio plans | Paid plans | Separate API | Not stated |
| Speechify API | 500,000 characters a month | Yes | Paid plans | Yes, also free | Yes | Not stated |
| HeyGen | 3 watermarked videos a month | Yes | One clone free | Paid plans remove branding | Yes | Not stated |
| Rask AI | 3 minutes | Yes | Yes | Not stated | Enterprise | Not stated |
| Cartesia | 20,000 credits a month | Yes | From Pro | From Pro | Yes | EU deployment, Enterprise |
| OpenAI | Billed per use | Yes, voices tuned for English | Eligible customers | Disclose the AI voice | Yes | Europe region, eligible customers |
| Google Cloud TTS | Free characters monthly, billing on | Yes | Allow-listed | Service terms | Yes | eu endpoint |
| Azure AI Speech | F0 tier | Yes | Limited access | Service terms | Yes | European regions |
| Amazon Polly | Free tier or credits | Yes | Brand Voice project | Service terms | Yes | AWS region you choose |
| NaturalReader | Free plan | Yes | Listed | Commercial product only | Not stated | Not stated |
| Chatterbox | Free, self-hosted | Yes | Yes | MIT license | Your own | Your hardware |
| Kokoro | Free, self-hosted | French, Spanish | No | Apache 2.0 | Your own | Your hardware |
| Piper | Free, self-hosted | Yes | No | Per voice license | Your own | Your hardware |
| XTTS-v2 | Free, self-hosted | Yes | Yes | Non-commercial only | Your own | Your hardware |
The Speechify API allowance for its newest model, Simba 3.5, drops to 250,000 characters a month after November 21, 2026. "Not stated" means we found no data-location statement on the pages listed in the sources.

Names you'll still see in older lists
Some ElevenLabs alternative lists haven't caught up with 2026:
- Resemble AI: on August 13, 2026, Resemble announced it no longer sells voice AI to new customers and now focuses on deepfake detection; existing customers are still supported. Its open models, including Chatterbox, stay free to build on.
- Descript: an editor for podcasts and videos, but not a way to leave ElevenLabs' technology: its help center lists the ElevenLabs v2, v3 and v4 models among the voice options for its AI Speakers, which need a consent statement read in English.
How to move your projects away from ElevenLabs
Plan the move before you cancel: ElevenLabs says unused credits expire at the end of the billing cycle when you downgrade or cancel. Content made during a paid subscription can still be used commercially afterward, subject to its terms.
- Make a list. Note every voice ID, model, pronunciation dictionary, Studio project and agent you use. Under Developers, the Analytics tab can export your usage view as a CSV file.
- Save your scripts. The History API returns each generated item with its text, voice ID, voice name, model and settings, up to 1,000 items per request. Music and sound effects aren't included.
- Download the audio you want to keep. From the history in Text to Speech you can download MP3 or WAV, and under Advanced also M4A and FLAC.
- Export your pronunciation dictionaries. Each dictionary version downloads as a PLS file.
- Map old voices to new ones. Keep a two-column list, old voice ID and new voice, so scripts can be regenerated in batches.
- Recreate clones from your original recordings. Clones can't be exported, and ElevenLabs notes that a new clone sounds slightly different even from the same audio. The new tool will also want its own consent recording.
- Run both for a while. Have a native speaker compare the same scripts in the old and new voice, then cancel.
This request fetches the first page of your history; repeat it with start_after_history_item_id set to the last ID while has_more is true:
curl -s "https://api.elevenlabs.io/v1/history?page_size=1000" \
-H "xi-api-key: $ELEVENLABS_API_KEY" \
-o history-page-1.jsonAnd this one downloads a dictionary version as a PLS file:
curl -s "https://api.elevenlabs.io/v1/pronunciation-dictionaries/DICTIONARY_ID/VERSION_ID/download" \
-H "xi-api-key: $ELEVENLABS_API_KEY" \
-o dictionary.plsKeep your API key in an environment variable, never in a script you share.
Troubleshooting after the switch
Names and brand terms sound wrong. Your dictionary didn't come along. Amazon Polly uses PLS lexicons too, so import yours and listen to each entry again. Cartesia expects a JSON list of words with an IPA or "sounds-like" pronunciation, so convert each entry. Google's Instant Custom Voice accepts custom pronunciations in IPA or X-SAMPA, still marked experimental.
Numbers and dates are read the wrong way. In German or Spanish, "1.234,56" and "25.10.2026" depend on the locale, so set the exact locale, not just the language. Cartesia normalizes dates and numbers for the locale you pass; Google, Azure and Polly voices are tied to locales such as es-ES or es-MX.
Pauses disappear or tags are read aloud. ElevenLabs' v3 and v4 models use audio tags and don't support SSML break tags, while Amazon Polly and the Speechify API use SSML and Cartesia supports a short list of SSML-like tags, such as break and speed. Convert or strip the tags before you regenerate.
A cloned voice loses its accent in another language. Cartesia advises recording the clone in a language the speaker actually speaks and adding other accents afterward. Descript notes that its clones are based on US English pronunciation.
The consent recording is rejected. Read the phrase exactly as shown, in a quiet room. Speechify transcribes the recording and refuses any change in wording; every refused attempt needs a new challenge and phrase. OpenAI also fails any recording that departs from its script.
The cloning endpoint returns an error 403. At Murf, cloning has to be switched on for your Enterprise workspace first. At Speechify, a 403 can mean cloning isn't available in your region.
Your free API stops mid-month. The Speechify API's free plan pauses until the next month and can't be topped up. Google Cloud starts billing once you pass the free characters, so set a budget alert when you enable billing.
If you only need a free voice for a one-off project, see our guide to free AI voice generators. For French voices in particular, our guide to French text to speech goes deeper.
