Female voice
ElevenLabs Eleven v4 · Jade
Never lost: 9 wins, 0 losses, 6 too close
- Just as good: ElevenLabs v4 Turbo · Talia (9–2–4), at half the credits
- Third: Cartesia Sonic 3.6 · Jacqueline (8–4–3)
The winners
Male voice
ElevenLabs Eleven v4 · Kaelen
Never lost: 11 wins, 0 losses, 7 too close
- Close behind: Cartesia Sonic 3.6 · Daniel (7–3–3)
- Close behind: ElevenLabs v4 Turbo · Darian (4–3–4)
Every sample on this page is an AI-generated voice, labeled with company, model and voice. Wins, losses and "too close" come from blind head-to-heads, both listeners together. "Too close" means the listener could not hear a winner. Two takes per voice: same voice, same settings, generated twice. Cartesia samples were re-recorded on a paid plan after the test, with the same voice and settings; the blind test used earlier takes made on Cartesia's free plan, which we may not publish.
What the test found
ElevenLabs Eleven v4 won both genders. Its two voices went 20 wins, 0 losses and 13 too close against every other voice. By company, ElevenLabs came first and Cartesia a clear second.
The cheaper ElevenLabs model is nearly as good. v4 Turbo "Talia" scored level with Jade. Turbo uses half the credits per character.
The same voice can sound different from take to take. Cartesia "Jacqueline" was clearly better on her second take. The winners were steady on both takes, but it pays to generate two and keep the better one.
An AI judge can't pick the winner yet. We also had an AI model score every clip. It separated good voices from weak ones, but agreed with our listeners on only 66% of close calls, so no ranking on this page comes from it.
Every voice, ranked
Blind head-to-heads on the 30-second ad, both listeners together, ranked by a standard paired-comparison model (Bradley-Terry). Only the marked voices are clearly ahead; below them the order is too close to call.
| # | Voice | Won–lost–too close | Result | Listen (first take) |
|---|---|---|---|---|
| 1 | ElevenLabs v4 · Jade | 9–0–6 | Picked best | |
| 2 | ElevenLabs v4 Turbo · Talia | 9–2–4 | Top group | |
| 3 | Cartesia Sonic 3.6 · Jacqueline | 8–4–3 | Top group | |
| 4 | Fish Audio Drama 3 · Hannah | 1–1–2 | Sample not published yet (license check) | |
| 5 | MiniMax Speech 2.8 HD · Upbeat Woman | 3–6–0 | Sample not published yet (license check) | |
| 6 | OpenAI gpt-realtime-2.1-mini · Marin | 1–5–3 |
Show all 12 female voices
| # | Voice | Won–lost–too close | Result | Listen (first take) |
|---|---|---|---|---|
| 7 | Google Gemini Flash · Kore | 3–6–0 | ||
| 8 | Google Gemini Flash-Lite · Kore | 0–4–2 | ||
| 9 | Murf Falcon 2.1 · Natalie | 1–2–0 | Sample not published yet (license check) | |
| 10 | Speechify Simba 3.5 · Brenda | 1–2–0 | ||
| 11 | Fish Audio s2.1-pro · Sheila | 0–2–0 | Sample not published yet (license check) | |
| 12 | Murf Gen2 · Samantha | 0–2–0 | Sample not published yet (license check) |
| # | Voice | Won–lost–too close | Result | Listen (first take) |
|---|---|---|---|---|
| 1 | ElevenLabs v4 · Kaelen | 11–0–7 | Picked best | |
| 2 | Cartesia Sonic 3.6 · Daniel | 7–3–3 | ||
| 3 | ElevenLabs v4 Turbo · Darian | 4–3–4 | ||
| 4 | Fish Audio Drama 3 · Ethan | 2–1–1 | Sample not published yet (license check) | |
| 5 | Google Gemini Flash-Lite · Charon | 1–3–7 | ||
| 6 | Speechify Simba 3.5 · Chase | 1–2–0 |
Show all 12 male voices
| # | Voice | Won–lost–too close | Result | Listen (first take) |
|---|---|---|---|---|
| 7 | OpenAI gpt-realtime-2.1-mini · Ash | 1–2–0 | ||
| 8 | MiniMax Speech 2.8 HD · Expressive Narrator | 1–2–0 | Sample not published yet (license check) | |
| 9 | Google Gemini Flash · Enceladus | 1–5–0 | ||
| 10 | Fish Audio s2.1-pro · Ethan | 0–2–1 | Sample not published yet (license check) | |
| 11 | Murf Falcon 2.1 · Wayne | 1–6–0 | Sample not published yet (license check) | |
| 12 | Murf Gen2 · Carter | 0–1–1 | Sample not published yet (license check) |
What one minute of voiceover costs
At API list prices, one minute of speech costs from about $0.009 (Google Gemini 3.8 Flash-Lite TTS) to $0.10 (MiniMax Speech 2.8 HD). Our winner, ElevenLabs Eleven v4, is about $0.08; v4 Turbo about $0.04.
One minute is about 1,000 characters; our clips used 800 to 1,000. Prices from each company's own pricing page; we keep a dated copy of every page we quote.
| Engine | List price | Per minute | Checked |
|---|---|---|---|
| ElevenLabs Eleven v4 (our winner) | $0.08 / 1,000 chars | $0.08 | |
| ElevenLabs Eleven v4 Turbo | $0.04 / 1,000 chars | $0.04 | |
| Cartesia Sonic 3.6 | Pro $5/mo for 100K credits | $0.05 | |
| OpenAI gpt-realtime-2.1-mini | per audio token | $0.024 | |
| Speechify Simba 3.5 | $0.02 / 1,000 chars | $0.02 | |
| Fish Audio Drama 3 / s2.1-pro | $0.05 / 1,000 chars / $0.015 / 1,000 chars | $0.05 / $0.015 | |
| Gemini 3.8 Flash / Flash-Lite TTS | per audio token | $0.014 / $0.009 | |
| Murf Falcon 2.1 / Gen2 | $0.01 / 1,000 chars / $0.03 / 1,000 chars | $0.01 / $0.03 | |
| MiniMax Speech 2.8 HD | $0.10 / 1,000 chars | $0.10 |
Can you use it commercially?
| Tool | Commercial use | Cheapest option for it | Checked |
|---|---|---|---|
| ElevenLabs | Paid plans only; the free plan is non-commercial | Starter, $6/month | |
| Cartesia | Paid plans only: Pro and up include a commercial license (ads included, by our reading of the terms); the free plan is non-commercial | Pro, $5/month | |
| Fish Audio | Paid plans; their own library voices are unclear (we asked) | Paid API | |
| Google Gemini | Yes, Google claims no ownership of the audio | Pay as you go (API) | |
| MiniMax | Yes on their own accounts (output is yours) | Pay as you go (API) | |
| Speechify | Yes, from the free API plan; disclose that the voice is AI | Free API plan | |
| OpenAI | Yes, you own the output; tell listeners the voice is AI | Pay as you go (API) | |
| Murf | Paid plans; the free Studio plan has no commercial rights | Studio from $19/month (annual) |
Questions people ask
What is the best AI voice generator for ads?
In our October 2026 blind test of 24 voices from 8 companies reading the same 30-second ad, ElevenLabs Eleven v4 was picked best: "Jade" (female) and "Kaelen" (male). Its two voices never lost a head-to-head: 20 wins, 0 losses, 13 too close to call.
Is anything better than ElevenLabs?
Not for ads in our test: no voice from another company beat ElevenLabs Eleven v4 in a blind head-to-head. The closest alternative was Cartesia Sonic 3.6, a clear second among the 8 companies.
Cartesia vs ElevenLabs: which sounds better?
For a 30-second ad, ElevenLabs sounded better to our blind listeners: in 8 direct duels ElevenLabs won 6, 2 were too close, and Cartesia won 0. Cartesia Sonic 3.6 was still a clear second of 8 companies. We did not test speed (latency), where the two are usually compared.
Is ElevenLabs v4 Turbo as good as Eleven v4?
Nearly: v4 Turbo "Talia" scored level with v4 "Jade" in our blind test (9–2–4 vs 9–0–6, too close to separate) and costs half as much per character.
Which AI voice sounds the most realistic?
The voices our blind listeners preferred for a real ad were ElevenLabs Eleven v4 "Jade" and "Kaelen", then v4 Turbo "Talia" and Cartesia Sonic 3.6. We asked "which would you use in an ad", which is close to, but not the same as, "which sounds most human".
How much does an AI voiceover cost per minute?
At API list prices, one minute of speech (about 1,000 characters) costs from about $0.009 (Google Gemini 3.8 Flash-Lite TTS) to $0.10 (MiniMax Speech 2.8 HD). The winner, ElevenLabs Eleven v4, is about $0.08; v4 Turbo about $0.04.
Is ElevenLabs free for commercial use?
No. ElevenLabs' free plan is for non-commercial use only; ads, client work and monetized videos need a paid plan (Starter or higher), according to ElevenLabs' terms checked on October 30, 2026.
Is PlayHT still available?
No. PlayHT (PlayAI) closed on December 31, 2025, after Meta took over its team in July 2025, and play.ht no longer resolved when we checked on October 30, 2026. Google's AI Overviews still recommended it in answers we collected on October 30, 2026.
How we tested
Same script for every tool, the vendor's newest model on the test day, voices chosen by a fixed rule, two takes per voice. Clips leveled to the same loudness, names hidden, sides shuffled per listener: two voices at a time, the listener picks the better one, until one is left. Every clip was transcribed by speech-to-text, and flagged spots were re-checked with 4 models.
The ad script every voice read:
Still wrapping gifts at midnight? Meet Novalo, the gift box that lights up the moment it opens. One tap, and your present feels like a little show. Order today for just fifty-nine ninety-five, and get free shipping across the US. Novalo. Make every gift a moment.
- Engines
- 12, from 8 companies
- Head-to-heads
- 92 (ranked script: 30-second ad)
- Voices
- 24 (12 per gender)
- Scripts
- ad, explainer, UGC, pronunciation, warm, phone menu
- Listening
- Blind, 2 listeners
- Loudness
- All at −16 LUFS
- Run
- Oct 26–30, 2026
- Tested by
- Nick
Independent. Rankings come from the test, never from commissions, and vendors don't see results before we publish. Disclosure
Changelog
- : First version: blind test results, prices, commercial licenses, questions.