Browser voice generators for a short narration: WellSaid, ElevenLabs, Cartesia and SpeechGen
WellSaid pairs consented actor voices with a contract that keeps your scripts out of model training, for English narration. Its lead is narrow. ElevenLabs leads independent listening results and costs less, but faces a voiceprint lawsuit and may train on your scripts unless you opt out.
WellSaid
Voices from actors who gave written consent and are paid when you use them, unlimited takes before you download, and a contract that keeps your scripts out of model training.
Its lead over ElevenLabs is narrow. Individual plans provide English voices and MP3 downloads; no independent listening test includes WellSaid.
Why WellSaid is our PickTeam behind WellSaid? Get your award badge
No Rising Star selected
Our October 29 assessment found model updates and browser generators, but no service established both a qualifying first release and suitable commercial narration rights.
Launch dates and commercial rights were unclear for several services; LazyVoice was Mac dictation rather than narration.
Who is this article for? People and small businesses turning an original short script into a narration they may publish commercially, using a ready-made voice in a desktop browser. Not voice cloning, dubbing or legal advice.
About 25 min read
What Preferred Pick, Rising Star and Ready to Rock mean
The quick list
Choose WellSaid for an original English script of one to five minutes and a file you may publish commercially. Compare the lowest commercial price, then how corrections are billed and where the voices came from.
US dollars before tax; not quotes. WellSaid and ElevenLabs prices checked October 29, 2026. We could not confirm every service’s current price.
The comparison at a glance
Compare the lowest commercial price, how corrections are billed, independent listening results, licences, voice consent and what happens to your scripts.
WellSaid is our choice for an original short English narration; its lead over ElevenLabs is narrow. Consider Cartesia for monthly delivery controls or SpeechGen for prepaid narration. All four remain selectable.
Hides matching answers for the selected tools. Unknowns and unverified claims stay visible.
| Feature | WellSaid | ElevenLabs | Cartesia | SpeechGen |
|---|---|---|---|---|
| Best fit | Preferred Pick: English narration with consented voices and no training on your scripts | Also consider: other languages, top-rated sound and a lower monthly price | Low monthly payment and delivery controls | Prepaid narration without a subscription |
| Lowest commercial plan | Starter US$19/month, or US$120/year | Starter US$6/month, or US$5/month billed yearly | Pro US$5/month | US$4.99/25K credits; US$24.99/200K for a Standard-voice year |
| Included on that plan | 20 downloaded minutes/month (240/year on yearly) | 30,000 credits, about 30 minutes of speech | 100,000 credits/month; one per character | 25K credits: 50K Standard, 25K Pro or 12.5K HD characters |
| Independent listening evidence | None found; not in the Artificial Analysis arena | Eleven v4 Turbo and v4 ranked 1st and 2nd of 97 (October 19, 2026) | Sonic 3.6 ranked fourth of 100; eight arena voices (October 29, 2026) | No independent comparison found for its Standard catalogue |
| Free corrections | Unlimited takes before download | Two regenerations of unchanged text within two hours; v3 charges every click | Changed generation uses credits; speed, volume and emotion controls | Cached unchanged sentences are free; changed sentences use credits |
| Commercial use | Full commercial rights on paid plans; you own inputs and outputs | Audio generated on a paid plan, indefinitely; not Beta services | Paid plans include commercial use; subject to source rights | Commercial use stated for free and paid audio; source rights remain yours to clear |
| Voice consent and legal record | Actors with written consent and residuals, per WellSaid; no adverse record found in October 19, 2026 search | Illinois voiceprint class action pending (filed May 2026); 2025 settlement with narrators | Stock-voice provenance not established; terms require user rights | Stock-voice provenance not established |
| Your scripts and data | Contract: no training on your inputs or outputs | May be used for training unless you opt out; opt-out covers new data only | May train on scripts/outputs; request an opt-out before submitting content | Contractual script-training and retention limits not established |
Original short scripts read by ready-made voices in a desktop browser and downloaded for commercial use. No voice cloning, real-time agents or dubbing. Capabilities and legal records checked October 19, 2026; pricing checked October 29, 2026; next review December 18, 2026, or earlier after a change in price, licence or legal status. This is public-source research: we created no accounts and generated no audio, so documentation and third-party listening results do not establish how a particular script will sound. WellSaid is our choice for English narration; its lead over ElevenLabs is narrow.
All features and sources 15 criteria
Open the groups that matter to you. They compare the same selected tools. “Planned” means future work, not an available feature. Sources include the published conditions.
Price and corrections
Show 7 criteria
| Feature | WellSaid | ElevenLabs | Cartesia | SpeechGen |
|---|---|---|---|---|
| Best fit | Preferred Pick: English narration with consented voices and no training on your scripts | Also consider: other languages, top-rated sound and a lower monthly price | Low monthly payment and delivery controls | Prepaid narration without a subscription |
| Lowest commercial plan | Starter US$19/month, or US$120/year | Starter US$6/month, or US$5/month billed yearly | Pro US$5/month | US$4.99/25K credits; US$24.99/200K for a Standard-voice year |
| Included on that plan | 20 downloaded minutes/month (240/year on yearly) | 30,000 credits, about 30 minutes of speech | 100,000 credits/month; one per character | 25K credits: 50K Standard, 25K Pro or 12.5K HD characters |
| Free evaluation | 3 download minutes/month; no commercial rights | 10,000 credits/month; no commercial use; must credit ElevenLabs | 500-character public previews; free account 20K/month, no commercial licence | Commercial use stated for free output; free allowance insufficient for ongoing job |
| What uses your allowance | Downloads only; generating takes is unlimited | Every generation; changed text is charged again | Characters generated; one credit each, break tags one credit | Characters generated: Standard 0.5, Pro 1, HD 2 credits each |
| Free corrections | Unlimited takes before download | Two regenerations of unchanged text within two hours; v3 charges every click | Changed generation uses credits; speed, volume and emotion controls | Cached unchanged sentences are free; changed sentences use credits |
| Unused allowance | Does not roll over | Rolls over up to two months while subscribed; lost on downgrade or cancellation | Rollover up to 2x monthly credits; paid credits retained on cancellation | One year; a top-up before expiry resets expiry for remaining credits |
These selected tools have matching answers in this group. Turn off “Show differences only” to see them.
Sources for price and corrections
Prices checked October 29, 2026: WellSaid pricing, ElevenLabs plans, Cartesia plans, SpeechGen packs and credit rules. US dollars before tax; future prices may change. WellSaid download rules and ElevenLabs regeneration rules last checked October 19, 2026.
Sound, languages and files
Show 4 criteria
| Feature | WellSaid | ElevenLabs | Cartesia | SpeechGen |
|---|---|---|---|---|
| Independent listening evidence | None found; not in the Artificial Analysis arena | Eleven v4 Turbo and v4 ranked 1st and 2nd of 97 (October 19, 2026) | Sonic 3.6 ranked fourth of 100; eight arena voices (October 29, 2026) | No independent comparison found for its Standard catalogue |
| Languages on self-serve plans | English only, several accents; others on Enterprise | Over 90 with Eleven v4; 29 with Multilingual v2 | 44 with Sonic 3.6, per Cartesia | 150+ languages, per SpeechGen; voice availability varies |
| Pronunciation and pacing | Respelling suggestions and a shareable pronunciation library | Audio tags in the script (v3, v4); v4 has no speed slider or SSML | Playground speed, volume and emotion controls | SSML and multi-speaker files; cached sentence corrections |
| Download formats | MP3 on Starter and Pro; SRT/VTT captions; 24 kHz on Starter | MP3, WAV; M4A and FLAC under Advanced | Browser project audio; MP3/WAV explicitly documented for API only | MP3, WAV, OGG |
These selected tools have matching answers in this group. Turn off “Show differences only” to see them.
Sources for sound, languages and files
Independent listening: Artificial Analysis, October 19, 2026 for ElevenLabs; October 29 for Cartesia. Cartesia browser FAQ and SpeechGen workflow checked October 29. WellSaid and ElevenLabs capabilities last checked October 19 except the October 29 WellSaid pricing FAQ correction to MP3-only individual plans.
Rights, consent and data
Show 4 criteria
| Feature | WellSaid | ElevenLabs | Cartesia | SpeechGen |
|---|---|---|---|---|
| Commercial use | Full commercial rights on paid plans; you own inputs and outputs | Audio generated on a paid plan, indefinitely; not Beta services | Paid plans include commercial use; subject to source rights | Commercial use stated for free and paid audio; source rights remain yours to clear |
| Voice consent and legal record | Actors with written consent and residuals, per WellSaid; no adverse record found in October 19, 2026 search | Illinois voiceprint class action pending (filed May 2026); 2025 settlement with narrators | Stock-voice provenance not established; terms require user rights | Stock-voice provenance not established |
| Your scripts and data | Contract: no training on your inputs or outputs | May be used for training unless you opt out; opt-out covers new data only | May train on scripts/outputs; request an opt-out before submitting content | Contractual script-training and retention limits not established |
| Main limitation | English and MP3 on individual plans; no independent listening test | Default voices end December 31, 2026; library voices can be withdrawn | Training use unless opted out; browser download format unspecified | No verified stock consent trail or independent Standard-voice listening result |
These selected tools have matching answers in this group. Turn off “Show differences only” to see them.
Sources for rights, consent and data
WellSaid agreement and ethics statement; ElevenLabs publishing rules, privacy and complaint, last adverse checks October 19, 2026. Cartesia terms and SpeechGen commercial permission, checked October 29. Complaints are allegations, not findings. No legal advice.
What matters when choosing
WellSaid is our Practical Pick for turning an original short script in English into a narration you can publish commercially, using a ready-made voice in a desktop browser. Its Starter plan costs US$19 a month for 20 minutes of downloaded audio, or US$120 billed yearly for 240 minutes, with full commercial rights. Generating and regenerating takes is unlimited on paid plans, so you pay only for the finished audio you download. WellSaid says every voice comes from an actor who gave written consent and is paid when customers use that voice, and that users have never been able to clone voices themselves. Its self-serve agreement says it will not use your scripts or audio to train its own or anyone else’s AI models. The most important compromise is evidence about sound: self-serve plans offer English voices only, and we found no independent listening test that includes WellSaid, so its sound rests on its own samples and mixed user reports. WellSaid pricing, download limit, ethics statement, self-serve agreement, languages by plan
Choose ElevenLabs instead if you need another language, want the voices with the strongest independent listening results, or want a lower monthly price. Its Starter plan costs US$6 a month for 30,000 credits, which ElevenLabs estimates at about 30 minutes of speech, with a commercial licence. In the Artificial Analysis speech arena, where listeners pick the more natural of two clips, its new Eleven v4 models ranked first and second of 97 models on October 19, 2026. Under our criteria, two things put it narrowly behind WellSaid for this reader. A May 2026 Illinois class action alleges ElevenLabs built its voice models from people’s voiceprints without consent. That is an allegation, not a finding. Its privacy policy also lets it train on your scripts unless you switch that off in your account. ElevenLabs plans, speech arena, complaint, May 11, 2026, privacy policy, October 9, 2026
Disclosure: Practical Picks articles are researched and written with AI, currently Anthropic’s Claude. ElevenLabs publishes an official connector that lets AI assistants, including Claude, generate speech with its service. We have no relationship with ElevenLabs or with Anthropic over this product, and the connector carried no weight in our judgment. ElevenLabs MCP server
Cartesia offers a low monthly payment and browser delivery controls; SpeechGen offers prepaid credits with no subscription. Neither gives us WellSaid’s clear combination of actor consent and a contractual no-training commitment. We have not selected a Rising Star: none of the newer services we could assess established a qualifying first release and suitable commercial narration service. Cartesia terms, SpeechGen, checked October 29, 2026.
Prices are in US dollars before tax; WellSaid and ElevenLabs prices were checked October 29, 2026. We could not confirm every service’s current price. WellSaid’s lead over ElevenLabs is narrow and rests partly on uncertain sound evidence; readers who care most about sound or price should choose ElevenLabs. Before paying for any of these services, try a passage with the names, abbreviations and numbers that matter most and note how each should sound. Pay before you generate the take you will publish: WellSaid, ElevenLabs and Cartesia require a suitable paid plan for commercial use; SpeechGen states that its commercial licence also covers free output. Save the final file and its plan details locally. This comparison covers ready-made voices only, not cloning a voice, real-time agents or dubbing. To edit a recording you already have, see our spoken-audio editor comparison.
WellSaid
Preferred Pick · English narration with consented voices
Choose WellSaid for an English narration when you want voices from consenting, paid actors, unlimited takes before you download and a contract that keeps your scripts out of model training.
Reasons to choose it
- Generating and regenerating takes is unlimited; only downloads count
- Voices built from actors who gave written consent and earn residuals, per WellSaid
- Agreement: no training on your inputs or outputs, and you own them
What to watch
- English voices and MP3 downloads on individual plans
- No independent listening test includes it
- Allowance does not roll over; no refunds after paying
WellSaid’s Starter plan costs US$19 billed monthly for 20 downloaded minutes a month, or US$10 a month billed yearly (US$120) for 240 minutes a year. Pro costs US$49 monthly for 180 minutes, or US$33 a month billed yearly. Both include full commercial rights and unlimited generation. The free trial includes 3 download minutes a month without commercial rights. Downloading a take or a combined file uses its length from your allowance; generating, regenerating, previewing and failed takes do not. Unused minutes do not roll over. Pricing, download limit, updated July 29, 2026
Choose WellSaid when you want to know where the voice came from and what happens to your script. WellSaid states that it does not train on open-source data, that its voices are built from private recordings by people who gave explicit written consent and receive residuals, and that users have never been able to clone voices independently. Its self-serve agreement, effective August 13, 2025, says you own your inputs and outputs and that WellSaid will not use them to train its own or any third party’s AI models. The agreement also provides for a custom-voice feature, which requires the speaker’s written consent. Choose ElevenLabs instead if you need a language other than English: self-serve plans include English voices in several regional accents, while Arabic, French, German, Japanese, Spanish and other languages require an Enterprise plan. Ethics statement, self-serve agreement, languages by plan
Read the full WellSaid analysis and sources
We found no independent listening comparison that includes WellSaid; it does not appear in the Artificial Analysis speech arena. User reports on Trustpilot describe its voices as realistic and professional, though one reviewer found them monotone without manual adjustment. Unlimited takes mean you can keep correcting a line without paying until you download. WellSaid’s help center documents dictionary-backed respelling suggestions in American and British English and a pronunciation library you can export and share, but we did not establish how well they work on a given name. Speech arena, Trustpilot profile, respelling suggestions, pronunciation library
WellSaid’s help center lists MP3, WAV and OGG as download formats, with optional SRT or VTT caption files. Its pricing page lists a 24 kHz sample rate for Starter and up to 48 kHz for Pro. The pricing FAQ checked October 29, 2026 limits Starter and Pro to MP3; Business and Enterprise offer additional formats. Choose another service if you need a WAV download on an individual plan. Download settings, pricing
The agreement also explains what you give up. WellSaid may review inputs and outputs, by people or automated tools, to enforce its acceptable-use rules. Payments are non-refundable except where the agreement or law says otherwise. A cancellation takes effect at the end of the current period, and you must cancel at least 24 hours before it ends to avoid the next charge. Disputes go to individual arbitration. Its privacy policy, last updated July 19, 2024, governs personal use; business users are covered by a data processing agreement. The contracting company is WellSaid Labs, Inc. We found no lawsuit, regulator action or security incident involving WellSaid in our October 19, 2026 search. Self-serve agreement, privacy policy
Used WellSaid? Share your experience to help other readers choose.
ElevenLabs
Also consider · other languages and top-rated sound
Choose ElevenLabs when you need a language other than English, want the voices that led independent blind listening results, or want a lower monthly price. Opt out of model training before you paste a script.
Reasons to choose it
- Eleven v4 models ranked first and second of 97 in a blind listening arena
- Over 90 languages with Eleven v4
- Paid-plan audio stays licensed for commercial use after you cancel
What to watch
- Pending Illinois class action alleges voiceprints were taken without consent
- May train on your scripts unless you opt out
- Default voices end December 31, 2026; library voices can be withdrawn
ElevenLabs’ Starter plan costs US$6 a month, or US$5 a month billed yearly, for 30,000 credits and a commercial licence. ElevenLabs estimates that as about 30 minutes of speech; its pricing FAQ gives text to speech at about one credit per character. Creator is US$22 a month, with the first month at US$11, for 121,000 credits. Unused paid credits roll over for up to two months while you stay subscribed, and are forfeited at the end of the cycle if you downgrade or cancel. The website allows up to 5,000 characters per generation on paid plans. Plans, text-to-speech guide
Choose ElevenLabs when sound or language coverage matters most. Its Eleven v4 model, released in September 2026, supports more than 90 languages and accepts inline directions such as pauses and whispers. In the Artificial Analysis speech arena on October 19, 2026, Eleven v4 Turbo and Eleven v4 held first and second place among 97 models, each from about 2,000 appearances in blind comparisons. That measures how natural listeners found short clips, not whether a particular name or number is read correctly. Check the voice before you settle on it. ElevenLabs’ original Default voices expire on December 31, 2026 and are available only to accounts created before March 2026. Many voices in its Voice Library are professional clones shared by their owners, who can withdraw them at once or after a notice period of 30 days to two years; some cost extra credits. Eleven v4, models, speech arena, Default voices, Voice Library, notice periods, paid-only voices and multipliers
Read the full ElevenLabs analysis and sources
Corrections cost credits. On the website, you get two free regenerations if the text, voice and model stay the same, the first take is less than two hours old and you have not refreshed or left the page; voice-setting sliders may change. Eleven v3 works differently: each click is charged and returns two alternatives. Changing the text always starts a new charged generation. Eleven v4 offers only Stability and Similarity settings, with no speed slider and no SSML; you direct delivery with audio tags in the script. ElevenLabs’ guide labels v3 “Alpha” and does not state a release status for v4. Regeneration rules, updated May 29, 2026, text-to-speech guide
The licence follows the plan you were on when you generated the audio. Audio made on a paid plan can be used commercially indefinitely, subject to ElevenLabs’ service-specific terms. Audio made on the free plan cannot be used commercially and must credit ElevenLabs when shared, even if you subscribe later. Beta services are excluded from commercial or production use, so check the model’s status. Downloads are MP3 or WAV, with higher-bitrate MP3, M4A and FLAC under Advanced in your history. Publishing rules, updated August 11, 2026, history downloads
Check the data setting before you paste a script. ElevenLabs’ privacy policy, updated October 9, 2026, says it uses personal information, which can include the text you submit and the audio it produces, to train its AI models. You can opt out under Data use in your account, but the opt-out covers only data submitted after you switch it off. Enterprise customers are not trained on by default. It keeps personal information while you have an account and data derived from your voice for up to three years after your last interaction. The policy names Eleven Labs Inc., with ElevenLabs Poland as controller for European users. Privacy policy, data-use setting
The legal record concerns how voices were collected and cloned, not ordinary stock narration. In Amer et al. v. Eleven Labs Inc. (N.D. Ill., 1:26-cv-05437, filed May 11, 2026), seven Illinois journalists and voice professionals allege that ElevenLabs captured voiceprints from their public recordings without the consent Illinois biometric law requires. These are allegations, not findings, and we could not open the court docket to confirm the case’s current status. An earlier suit by audiobook narrators and authors ended with a stipulated dismissal with prejudice after settlement in November 2025. In March 2025, Consumer Reports found that ElevenLabs’ instant cloning relied on a checkbox confirmation of rights; ElevenLabs pointed to voice verification for professional clones and to blocks on hundreds of public figures’ voices. In April 2026, Senator Maggie Hassan asked ElevenLabs and three other companies how they stop scammers; her office alleged no wrongdoing. Complaint, Vacker dismissal, November 6, 2025, Consumer Reports report, Senator Hassan’s letters
Used ElevenLabs? Share your experience to help other readers choose.
Cartesia
Low monthly payment and delivery controls
Consider Cartesia for low monthly spending and browser speed, volume and emotion controls. Request its training opt-out before submitting a script.
Reasons to choose it
- Monthly commercial plan with no annual commitment
- Sonic 3.6 ranked fourth of 100 in independent blind listening results on October 29
- Overages can be turned off
What to watch
- May train on scripts and outputs until an opt-out is processed
- Stock-voice provenance not established
- Browser download format unspecified; MP3/WAV named for API only
Cartesia Pro costs US$5 monthly for 100,000 credits and a commercial-use licence. Sonic text to speech uses one credit per character, so our allowance of 27,000 generated characters a month fits without buying extras. The free plan has 20,000 credits and does not include commercial permission. Its public voice generator is a listening preview; Cartesia directs project audio to its browser Playground, where it documents speed, volume and emotion controls. Its FAQ names MP3 and WAV for the API, but does not specify the browser download format. Pricing, voice generator and Playground FAQ, checked October 29, 2026.
Choose Cartesia if you want a low monthly payment and control over delivery. Sonic 3.6 ranked fourth of 100 models in Artificial Analysis’s blind listening arena on October 29, with 3,126 comparisons across eight voices. That supports its sound, but does not tell you how your chosen voice will pronounce a name. Its terms allow Cartesia to use scripts and outputs for model training unless you request an opt-out; that request applies after processing and does not undo earlier use. The terms name Cartesia AI, Inc., but do not establish the provenance of each ready-made voice. WellSaid offers a clearer actor-consent and no-training commitment. Cartesia’s pricing lets you turn overages off; do so if you want generation to stop at your balance. Speech arena, terms, revised June 14, 2024, checked October 29, 2026.
Used Cartesia? Share your experience to help other readers choose.
SpeechGen
Prepaid narration without a subscription
Consider SpeechGen when you prefer prepaid credits. Standard voices cost half a credit per character; Pro and HD voices cost more. Check the voice before committing to a script.
Reasons to choose it
- No subscription or automatic renewal
- Unchanged cached sentences cost no credits to repeat
- SSML, multi-speaker files and MP3/WAV/OGG downloads
What to watch
- Credits expire after one year unless topped up
- Stock-voice consent and script-training limits not established
- No independent listening comparison of its Standard catalogue
SpeechGen sells credit packs with commercial permission rather than a subscription. Standard voices use one credit per two characters, Pro voices one per character and HD voices two per character. US$24.99 buys 200,000 credits, enough for a year of our ten finished minutes a month with up to three attempts per script using Standard voices. Credits last one year; topping up before expiry carries the remaining balance forward and resets its expiry. Higher voice tiers use the same balance faster. Pack prices and credit rules, checked October 29, 2026.
Choose SpeechGen when you prefer paying as needed and want SSML, several voices in one file and MP3, WAV or OGG downloads. It caches generated sentences, so unchanged sentences do not cost credits again. Its pricing page grants commercial use, but we could not establish the rights chain for each ready-made voice or contractual limits on using your scripts for training. We found no independent listening comparison of its Standard catalogue. Try the particular voice before committing to a long script; a large catalogue does not establish consistent pronunciation or consent. SpeechGen workflow, pack details, checked October 29, 2026.
Used SpeechGen? Share your experience to help other readers choose.
Which product are we missing?
Know a product that belongs in this article? Tell us what it does differently and why it deserves a closer look. Newcomers welcome.
Readers and product teams can suggest candidates. Suggestions inform our research; they do not guarantee inclusion or a recommendation.
Opens an email draft with this article attached as a link. Nothing is sent until you send it.
No email app? Send the product name, website, and this article’s title to admin@practicalpicks.guide.
What people are saying
User reviews are a check on vendor claims, not a measured reliability score. We read the most recent reviews on each Trustpilot profile on October 19, 2026 and report the themes, not star averages.
ElevenLabs’ recent reviews, from October 7 to 18, 2026, concentrate on billing and credits: unexpected recurring charges, difficult refunds, credits used faster than expected and an account suspended with credits left. One reviewer found the voices unrealistic for long documents. A Reddit discussion we read on September 7, 2026 describes inconsistent pronunciation-dictionary behavior with v3 through the API. WellSaid’s profile is small, with 23 reviews: recent reviewers praise realistic voices and responsive staff, while others report renewal charges after cancelling, refused refunds and slow support. ElevenLabs reviews, v3 dictionary discussion, WellSaid reviews
Narakeet’s 15 reviews were not invited by the company. Reviewers like its free voice picker for short phone messages; complaints concern how quickly the free allowance runs out, and one September 2026 reviewer says voices tried in the free trial were locked after a paid purchase, which we could not verify. Murf reviewers praise its voices and pronunciation options; complaints describe a laggy editor, automatic renewals, refused refunds and, in September 2026, cancellation requests that went unanswered. A 2024 instructional-design discussion reports enterprise renewal friction. Speechify’s profile is dominated by recent reviews praising its support, mostly about the reading app rather than Studio. Narakeet reviews, Murf reviews, Murf pricing discussion, Speechify reviews
How we researched
This is public-source research. We did not create accounts, sign in, buy plans, install software, generate or listen to audio, or contact the companies. No company supplied access or payment for this work, and no link carries affiliate tracking. We first compared ElevenLabs, Murf Studio and Narakeet on September 7, 2026. On October 19, 2026 we rechecked prices, plan limits, correction charges, licences, data terms and legal records, and added WellSaid and Speechify Studio after checking which widely used browser voice generators the comparison was missing. On October 29, 2026 we checked public pricing and newer voice services again; Murf’s page showed API pricing rather than its Studio plans. ElevenLabs’ help pages blocked our automated requests, so we read the same published articles through the help center’s public data feed.
We judged each service for a person or small business with one job: an original short script, one to five minutes long, read by a ready-made voice and downloaded as a file you may publish commercially. We fixed five criteria and their weights before scoring: sound and correction control (30%), the licence for that use (20%), how the voices were obtained and the consent record (15%), cost for about ten finished minutes a month over three years (20%), and treatment of your scripts and data (15%). Sound evidence comes from the Artificial Analysis speech arena, a public leaderboard built from blind listener votes between pairs of clips. It is the only independent comparison we found that covers several of these services. WellSaid scored 3.81 out of five and ElevenLabs 3.70; its lead is narrow. Cartesia and SpeechGen both round to 3.47, with Cartesia slightly ahead before rounding. Voice-consent, data and legal findings for the original services were last checked October 19; pricing and the additional alternatives were checked October 29.
We excluded tools that no longer serve new customers. LOVO’s operator filed for Chapter 7 bankruptcy on May 27, 2026, and its website was not serving on October 19. We reconsidered editors with voice features when they could also produce a standalone narration: an editor’s main purpose alone is not a reason to exclude it. Lovo notice of bankruptcy
On October 29 we checked product announcements, release histories and the Artificial Analysis speech arena for newer browser voice services. Directory pages were unavailable. Fish Audio’s paid-service terms allow commercial use, but its pricing FAQ restricts that use to verified voices you own, which does not establish permission for the ready-made voices this comparison covers. MiniMax’s commercial download terms were unavailable, Hume’s pricing link led to a voice-AI evaluation service, and LazyVoice was Mac dictation rather than narration. Model release dates alone do not establish a new browser service. No newer service met the requirements for our Rising Star recommendation. Fish Audio terms, plans, MiniMax Audio, Hume, LazyVoice, checked October 29, 2026.
Narakeet remains useful for occasional narration: US$6 buys 30 built minutes with no subscription or expiry, and paid accounts can preview without a charge. Its larger-pack prices were unavailable, so we could not establish the cheapest cost for regular use. Murf Studio suits repeated delivery changes to unchanged text, which consume no generation allowance; its current Studio price was unavailable. Murf’s commercial rights cover original monetized videos, but TV or syndicated radio requires Enterprise and it recommends a business licence for registered businesses. Murf says every Studio voice is available with the consent of the voice actor and that it retires a voice, with notice, if the actor ends the collaboration. Its terms give Murf only a licence to process your scripts, which ends when you delete them or close your account; its privacy policy says projects and uploaded text are deleted within 30 days of termination, with legal and security exceptions, and neither document says whether your content is used to train models. These terms, and our searches for lawsuits or regulator action against either service, were last checked October 19. Narakeet pricing, Murf generation-time rules, commercial rights, retired voices, terms, section 3, privacy policy.
Typecast scored 2.65 out of five for this audience. Its pricing says businesses, institutions and agencies should use Business: US$69 monthly or US$744 yearly, with 200,000 download credits a month. Basic is US$54 yearly with 30,000 credits for a personal customer; it is not the small-business price. Listnr Individual scored 2.69 and costs US$190 yearly with about two hours of generation a month and commercial permission. NaturalReader Commercial and Fliki document relevant workflows, but we could not confirm a suitable voice-credit rate or plan price. Descript documents stock voices and standalone audio exports, but its commercial terms and speech-credit rate were unavailable. Clipchamp, CapCut, Canva, VEED, TTSMaker, Voicemaker and Resemble lacked enough information about the relevant download, licence or plan for a complete comparison. Typecast, Listnr, NaturalReader, Fliki, Descript, Resemble, checked October 29, 2026.
Speechify Studio scored 2.73 out of five and remains outside the main comparison. Studio Starter costs US$100 yearly for about 24 hours of voiceover, but requires yearly payment; its terms grant a broad, permanent licence to submitted material. Its cloning safeguards also face a consumer-group complaint. The complaint is an allegation, not a regulator finding. Studio plans, terms, checked October 29, 2026; CFA complaint, checked October 19, 2026.
Our adverse search through October 19 covered lawsuits, insolvency filings, regulator actions, consent complaints and security reports for each incumbent. We did not repeat those independent adverse checks on October 29 for the additional services. A search that finds nothing is not proof that nothing exists.
| Tool | Public sources | What remains uncertain |
|---|---|---|
| WellSaid | Pricing page, self-serve agreement (August 2025), ethics statement, help articles on downloads and languages, privacy policy, user reviews. | How its voices compare with others in a blind test and how well pronunciation tools work on a given word. Individual plans provide MP3, per the October 29 pricing FAQ. |
| ElevenLabs | Pricing, model and voice documentation, help articles on regeneration, licensing, downloads and voices, October 2026 privacy policy, court filings, Consumer Reports, Senate letter, speech arena. | Status of the Illinois case, the website credit rate and commercial release status for Eleven v4, whether a given library voice will stay available. |
| Cartesia | October 29 pricing, browser Playground FAQ, terms and independent speech arena. | Browser download encoding, stock-voice consent trail and handling of a particular script. Training opt-out is prospective. |
| SpeechGen | October 29 browser workflow and pack prices, commercial licence, voice-tier rates and expiry rules. | Stock-voice provenance, contractual script-training limits and sound of a particular Standard voice. |
| Narakeet and Murf | October 19 rights, controls, data terms, user reports and adverse checks; October 29 pricing attempts. | Cheapest Narakeet larger pack and current Murf Studio price. |
| Newer services and other alternatives | October 29 pricing pages, model leaderboard and release histories; October 19 adverse records. | First browser-product launch dates, suitable prices and commercial download terms for several services; no complete new adverse search. |
None of this establishes how a particular script will sound, whether a name will be pronounced correctly, or how a service handles your data in practice. Next substantive review is due December 18, 2026, before ElevenLabs’ Default voices expire, or sooner if a price, licence or legal status changes. Found a changed price or term? Contact the editors.
Sources and evidence
Sources were checked October 19, 2026 unless another date is given. Pricing pages and newcomer links explicitly dated October 29, 2026 were checked on that date; this was not a full refresh of every claim. Company material establishes documented offers and terms, not performance. Court complaints contain allegations, not findings. Review-platform and community posts are attributed experiences, not representative samples.
- WellSaid (WellSaid Labs, Inc.): pricing; self-serve services agreement, effective August 13, 2025; ethics statement; privacy policy, July 19, 2024; help articles, read October 19, 2026, on download limits, download formats, languages by plan and respelling suggestions, and on October 12, 2026 on the pronunciation library.
- ElevenLabs (Eleven Labs Inc.): plans; text-to-speech guide; models; Eleven v4; privacy policy, October 9, 2026; data-use setting; MCP server. Help articles, read through the help center’s public data feed because the pages blocked automated requests: regeneration, May 29, 2026; publishing, August 11, 2026; history downloads, December 31, 2025; Default voices, notice periods and paid-only voices, August 14, 2026; Voice Library, March 20, 2026.
- Narakeet (Video Puppet Limited): pricing; usage rights; formatting reference; data security, October 31, 2025; privacy policy; text-to-audio converter, read September 7; Companies House record.
- Murf: generation time; validity; carry-forward terms; free-trial downloads; commercial rights; retired voices; products; voice-plus-music export; mobile web app; terms; privacy policy; pricing page, checked October 29, 2026, showing API rates rather than Studio prices.
- Speechify (Speechify Inc.): Studio pricing and FAQ; terms, July 1, 2025; privacy policy, July 22, 2026.
- Independent listening evidence: Artificial Analysis speech arena, blind pairwise votes, ranks and release dates as displayed October 19, 2026 for the original services and October 29, 2026 for Cartesia.
- Legal and regulatory records: Amer et al. v. Eleven Labs Inc., complaint, N.D. Ill. 1:26-cv-05437, May 11, 2026; Rogers et al. v. Amazon, complaint, 1:26-cv-05497, May 12, 2026; complaint against Microsoft, 1:26-cv-05491, May 12, 2026; Vacker v. Eleven Labs, D. Del. 24-987, stipulated dismissal with prejudice, November 6, 2025, through a public mirror; Lehrman v. Lovo, S.D.N.Y. 1:24-cv-03770, notice of bankruptcy, May 27, 2026, through a public mirror; Senator Hassan’s letters, April 16, 2026; Consumer Federation of America complaint announcement, July 27, 2026; Consumer Reports, AI voice cloning report, March 10, 2025.
- User reports, read October 19, 2026 unless noted: Trustpilot ElevenLabs, WellSaid, Narakeet, Murf, Speechify; Reddit, read September 7, 2026: ElevenLabs v3 dictionary, Murf pricing, September 2024.
This is a researched comparison, not a product review. Method: evidence standard, editorial picks.