AI voice generators
AI voice generators turn text into natural speech and can clone or design voices. The market splits three ways: developer platforms sold by the character, creator apps sold by the month, and open-weights models you run yourself for free. This grid compares popular products side by side on what decides a choice: price and allowance, languages and voices, cloning and emotion control, speed for live conversation, commercial rights and consent safeguards. The real trade-offs are quality and polish against cost, and convenience against control.
Last updated
How to choose
Start with how you will use the voice, then check rights and cost per hour of audio before comparing how natural it sounds.
1. Decide what you need
- Narration, videos and podcasts: a creator app with a good voice library and Emotion and style control is usually easiest.
- An app or phone agent: you need an API, Real time streaming and low Streaming latency.
- Your own or a brand voice: check Instant voice cloning, Professional voice cloning and Voice consent verification.
- Many languages: compare Languages and Dubbing and translation.
2. Rule out deal-breakers
- Business use: check Commercial use; free tiers often forbid it, and some open-weights models are non-commercial only.
- Need to run it yourself, offline, or avoid per-character fees: only Open weights models qualify, and you pay in hardware and setup.
- Cloning a person's voice needs their consent: prefer tools that verify it.
3. Compare the real cost
- Plans count characters, credits or minutes, which are not comparable. Convert to the amount of audio you will actually make.
- Developers should compare API price per million characters for standard voices; premium voices cost more.
- Check whether the cheapest plan allows commercial use and cloning.
4. Judge quality fairly
- Demos use the best voice and script. Test the same three passages, including names, numbers and an emotional line, in two or three tools.
- The Independent Elo score comes from public blind voting and is one outside measure of quality.
5. Before you commit
- Voice companies are acquired or shut down often, and models are retired on a schedule. Keep your scripts and exports, and avoid long prepaid plans.
Answer
Browse the table. Filter to what matters.
Use the filters below, then sign in if you want to shortlist and decide with friends.
Sign in to vote, hide services and features, and reorder.
Verified: hover a cell and click its green dot for the source Estimate: no cited source Not stated: the company doesn't publish it
| Feature | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ | ⠿ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
⠿
Free plan
|
|
|
|
|
|
|
|||||||||||||||||||
|
⠿
Free plan limits
|
Free to download and run yourself; Apache 2.0 allows commercial use; you pay for your own hardware
|
Doesn't apply
|
10 minutes of voice generation in total, no downloads, no commercial rights
|
3 download minutes per month, no commercial rights, 3 projects
|
|||||||||||||||||||||
|
⠿
Cheapest paid plan
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
$29/month
|
$19/month
|
Doesn't apply
|
||||||||||||
|
⠿
Top individual plan
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
$99/month
|
$49/month
|
Doesn't apply
|
||||||||||||
|
⠿
Plan allowance
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
2 hours of voice generation per month, 100 projects (Creator)
|
20 downloaded minutes per month (unlimited generation)
|
30,000 TTS characters per month (TTS only plan)
|
Doesn't apply
|
|||||||||||
|
⠿
API price per million characters
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
$10/1M characters
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
||||||||||||||||||
|
⠿
Languages
|
41 languages
|
1 languages
|
6 languages
|
||||||||||||||||||||||
|
⠿
Voice library
|
None (zero-shot cloning and voice design only)
|
None (zero-shot cloning and voice design only)
|
36 English Flux TTS voices (Aura-2 adds more in 7 languages)
|
332 system voices
|
|||||||||||||||||||||
|
⠿
Instant voice cloning
|
|
|
|||||||||||||||||||||||
|
⠿
Professional voice cloning
|
|
|
|
|
|
|
|
|
|
|
|
|
|||||||||||||
|
⠿
Voice design from a prompt
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||
|
⠿
Emotion and style control
|
|
|
|||||||||||||||||||||||
|
⠿
Real time streaming
|
|
|
|
|
|
|
|||||||||||||||||||
|
⠿
Streaming latency
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
Doesn't apply
|
|||||||||||||||||||
|
⠿
Dubbing and translation
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||
|
⠿
Speech to text
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||||
|
⠿
Commercial use
|
Yes on all plans
|
Yes on all plans
|
Yes on all plans
|
Yes on all plans
|
Yes on all plans
|
Yes on all plans
|
Yes on paid plans
|
Yes on paid plans
|
Yes on paid plans
|
Yes on all plans
|
|||||||||||||||
|
⠿
Voice consent verification
|
|
|
|
|
|
|
|
|
|
|
Doesn't apply
|
|
|
|
|||||||||||
|
⠿
API available
|
|
|
|
|
|
||||||||||||||||||||
|
⠿
Open weights
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|||||||||
|
⠿
SSML support
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|||||||||||
|
⠿
Audio watermarking
|
|
|
|
|
|
||||||||||||||||||||
|
⠿
Independent Elo score
|
1,338 points
|
1,150 points
|
1,270 points
|
1,181 points
|
Why these features matter
-
Free plan
Free access is the easiest way to test voice quality.
-
Free plan limits
Free tiers often cap volume and forbid commercial use.
- Free to run yourself
- : The model weights are free, but you provide the computer and the setup.
-
Cheapest paid plan ($/month)
The entry cost for regular use.
-
Top individual plan ($/month)
High volume and best quality often need a higher tier.
-
Plan allowance
Characters, credits and minutes are not comparable until you convert them.
- Unlimited generation
- : You pay for downloads or minutes of final audio, not for every attempt.
- Credits
- : Credits are each maker's own unit; how far they go depends on the voice model and text length, so compare the stated minutes or characters.
-
API price per million characters ($/1M characters)
The cost at scale for apps and audiobooks.
-
Languages (languages)
Global content needs many languages.
-
Voice library
A bigger library makes it easier to find the right voice.
- No preset voices
- : You supply a short recording or describe a voice each time instead of picking from a menu.
-
Instant voice cloning
Quick custom voices for creators and brands.
-
Professional voice cloning
Closer match for brand and narration voices.
-
Voice design from a prompt
Makes unique voices without a recording or a real person.
-
Emotion and style control
Natural, expressive narration needs delivery control.
-
Real time streaming
Required for voice agents and live assistants.
-
Streaming latency (ms)
Lower latency feels like a natural conversation.
-
Dubbing and translation
Reaches new audiences without re-recording.
-
Speech to text
One vendor for both directions.
-
Commercial use
Business use needs clear rights.
- Yes on paid plans
- : Free or trial output is personal use only; you need a paid plan to publish or sell what you generate.
- Limited
- : The license restricts some commercial uses, so read it before building a business on this model.
-
Voice consent verification
Protects against misuse and legal risk.
-
API available
Needed to build apps.
-
Open weights
Lets you run it locally with no per-character cost.
-
SSML support
Fine control of pronunciation and pacing.
-
Audio watermarking
Helps identify AI-made audio.
-
Independent Elo score (points)
An independent, vote-based quality score.
About the services
-
Kokoro 82M · hexgrad
Tiny 82M-parameter Apache model that runs on a laptop CPU with 54 preset voices in 8 languages. Best for cheap, fast narration; it has no voice cloning.
-
Qwen3-TTS · Alibaba Cloud
Apache-licensed family from Alibaba with separate models for preset voices, 3-second cloning and voice design across 10 languages, plus streaming. Cloning and design need different checkpoints.
-
Fish Audio S2 Pro · Fish Audio · Superseded Jun 2026
Fish Audio's previous-generation S2 model: 80+ languages, free-form inline emotion tags, about 100 ms to first audio on an H200. Fish Audio now recommends S2.1 Pro, which keeps the same bracket-tag control and adds 83 languages. Weights are research and non-commercial only; commercial use needs a paid licence.
-
OmniVoice · k2-fsa
Zero-shot model covering 600+ languages with cloning and attribute-based voice design, very fast (RTF 0.025). Weights are CC-BY-NC, so no commercial use.
-
VoxCPM2 · OpenBMB
2B Apache model with 30 languages, 48kHz output, cloning, voice design and streaming; free for commercial use. Needs a GPU for best speed.
-
Supertonic 3 · Supertone · No longer supported Aug 2026
Very fast on-device model in 31 languages with laugh and breath tags, shipped with preset voice styles under an OpenRAIL-M license. Custom voices go through a separate builder tool.
-
Breeze TTS 2 · BreezeBlue
Highest-scoring open-weights model on Artificial Analysis with cloning, voice design and under 40 ms time to first audio, but English and Chinese only and non-commercial weights.
-
ElevenLabs · ElevenLabs
Most expressive and widest-ranging platform here: Eleven v4 TTS in 90+ languages, a 17,000+ voice library, instant and professional cloning, voice design, dubbing and speech to text, from a $6 plan. Credits are shared across products, and the free plan is non-commercial.
-
OpenAI text to speech (gpt-4o-mini-tts) · OpenAI · Superseded Oct 2026
OpenAI’s GPT-4o mini TTS API model is deprecated and scheduled to shut down January 6, 2027. OpenAI recommends migrating to gpt-realtime-2.1-mini through its Realtime API, a different workflow.
-
Google Cloud Text-to-Speech / Gemini TTS · Google
Developer API with a 4-million-character monthly free tier on Standard voices, 380+ voices in 75+ languages, SSML, prompt-steerable Gemini TTS and Chirp 3 instant custom voice. Newest Gemini 3.8 TTS models are in preview and voice design is on the Gemini Enterprise API only; billing must be enabled.
-
Azure AI Speech · Microsoft
Enterprise speech service covering text to speech, transcription and translation in one resource, with 100+ languages, HD emotion-aware voices, SSML, and gated personal and professional custom voices. Custom voice needs an access application; pay-as-you-go pricing only.
-
Amazon Polly · Amazon
Low-cost AWS API ($4 per million characters for Standard voices) with SSML, lexicons and a streamable Generative engine. No self-serve voice cloning (custom Brand Voice is built with AWS), and its free tier lasts only 12 months for new AWS accounts.
-
Cartesia Sonic · Cartesia
Low-latency Sonic 3.6 TTS (first audio under 90 ms) in 44 languages with 500+ voices, 10-second instant cloning and a $5 Pro plan; the free plan has 20K credits and no commercial license. Built for voice agents; tags are SSML-like rather than full SSML.
-
Deepgram Aura / Flux TTS · Deepgram
Voice-agent TTS (Flux TTS, Aura-2, Aura-1) priced per character with no subscription and a one-time $200 credit, plus speech to text on the same API. No voice cloning or design, and no SSML.
-
Inworld Realtime TTS-2 · Inworld
Realtime TTS-2 and Flash: steerable by plain-language direction, 500+ languages, cloning from 3 seconds, voice design, and per-character rates from $15 on demand down to $7 on the $1,500 plan. Voice design is in preview and cloning checks only a rights checkbox.
-
MiniMax Speech 2.8 · MiniMax
Speech 2.8 HD and Turbo with 40 languages, 7 emotions, 10-second cloning and voice design, from a $5 Audio plan or pay-as-you-go API at $60-100 per million characters. Cloned voices expire if unused for 7 days; designed or cloned voices cost $1.50-$3 each.
-
Murf · Murf AI
Studio-style voiceover editor for presentations, e-learning and ads, plus the Falcon API for voice agents. Cloning is sales-led rather than self-serve, and the free plan cannot download audio.
-
WellSaid Labs · WellSaid Labs
Voiceover studio built on licensed voice-actor recordings, aimed at corporate training, marketing and ads. Self-serve plans are English-only, there is no do-it-yourself voice cloning, and plans limit downloaded minutes rather than generation.
-
Speechify · Speechify
Best known as a read-aloud app (Premium) with a Studio for voiceovers and dubbing, and a separate developer API (SpeechifyAI) with the Simba model. App and API are billed separately, so check which one you need.
-
Fish Audio · Fish Audio
Low-cost voice generator and API with large community voice library, emotion tags, voice design and cloning. The free plan is non-commercial and capped at 7 minutes a month.
-
Respeecher · Respeecher
Voice conversion (speech-to-speech) and text-to-speech marketplace with licensed celebrity and talent voices, used in film and games. Credits-based pricing with no free plan, only a 3-day trial; cloning your own voice needs a custom plan.
-
Descript · Descript
Text-based video and podcast editor with built-in AI speech and voice clones. Suits creators who want to fix or add narration while editing; TTS is a feature, not the whole product, and plans are sized in media hours.
-
Voice.ai · Voice.ai
Cheap credit-based text to speech, 15-second voice cloning, voice changer and voice-agent platform. Free plan allows no instant clones and 500 characters per conversion.
-
NaturalReader · NaturalSoft Ltd.
Commercial voiceover plans that bundle Gemini, OpenAI, Azure and ElevenLabs voices with prompt-controlled delivery and voice cloning. Free use cannot download audio; prices shown are annual-billing monthly equivalents.
-
Chatterbox · Resemble AI · No longer supported Aug 2026
MIT-licensed zero-shot cloning model with an emotion exaggeration dial; Multilingual V3 covers 23 languages and output carries a Perth watermark. Resemble AI retired it in August 2026: weights remain downloadable, but there is no maintenance and no hosted API.
Suggest a change
Related comparisons
-
AI image generatorsAI image generators turn a text prompt or a reference picture into a finished image. This grid compares popular tools side by side on what decides a choice: what you can do for free, what the paid plans really include, which model powers each one, whether it edits and keeps characters consistent, and what rights you get to use the results commercially. The real trade-offs are quality and control against price, and ease of use against how open and legally safe the output is.
-
AI video generatorsAI video tools turn a prompt, a picture or a script into a finished clip, and the field splits into three kinds: general text-to-video models, avatar presenters that read your script, and AI editors that cut and polish footage you already have. This grid compares popular tools side by side on what decides a choice: what is free, what the paid plans really include, how long and sharp the clips are, whether they come with sound, and what rights you get. The real trade-offs are quality against credits (video burns credits fast) and creative control against ease of use.
-
AI meeting assistantsAI meeting assistants listen to your calls, write the transcript, summarise the discussion and pull out action items so you can pay attention instead of typing. This grid compares popular tools side by side on what decides a choice: how they capture the meeting (a visible bot, a quiet device recording, or the meeting platform itself), what the free plan covers, what paid plans cost per person, and how your meeting content is protected. The real trade-offs are convenience against privacy, and a cheap personal tool against a platform-wide one your company can control.
-
AI coding toolsAI coding tools now range from autocomplete in your current editor to agents that work on tasks in the cloud while you do something else. This grid compares popular tools side by side on the practical things that decide a purchase: where they run, what a plan really costs, which models and agent features you get, and how your code is handled. The real trade-off is power against predictability and control: the most capable agents cost more, often bill by usage, and need more care over privacy and review.
-
AI chatbotsAI chatbots now do much more than answer questions: they search the web, write reports, create images, talk, remember you and carry out tasks. This grid compares popular chatbots side by side on what decides a choice: what you get free, what each paid plan costs, which models and tools are included, and how your chats are treated. The real trade-off is capability against price and privacy: the strongest features sit on higher-priced plans, and many chatbots use your conversations to improve their models unless you opt out.
-
AI search enginesAI search tools read the web for you and answer in a few paragraphs with links, instead of giving you ten blue links. This grid compares popular AI answer engines, search engines that now add AI answers, and the developer APIs that power AI apps, on what decides a choice: what is free, what paid plans cost, whether answers cite sources, how deep the research goes, who runs the search index, and what happens to your queries. The real trade-offs are convenience against privacy, and speed against how much you can trust and verify an answer.