"AI audio" now covers at least four different jobs: turning text into natural speech, powering real-time voice agents, composing full songs, and generating sound effects. The best tool for one is rarely the best for another, and the pricing models differ enough that the cheapest option in one category can be the most expensive in the next.
This guide covers the tools worth your time in each category, what they actually cost, and the non-obvious details that decide which one you should pick.
Every price below was checked against each vendor's own pricing page on 18–19 September 2026. We have no commercial relationship with any of the companies listed.
The quick picks
| If you need | Pick | Why |
|---|---|---|
| One tool for everything | ElevenLabs | Speech, dubbing, music, sound effects and agents on one account |
| The cheapest quality voice | Fish Audio | Several times cheaper per minute than the market leader on annual plans |
| Real-time voice agents | Cartesia | Built for live conversation, with the lowest per-minute price here |
| Expressive voices and cloning | Hume | Unlimited voice cloning on every plan, including free |
| Complete songs with vocals | Suno | The fastest route from a prompt to a finished track |
| Detailed music editing | Udio | Longer generations and finer creative control |
| Licensed-safe music and sound effects | ElevenLabs Music and Sound Effects | Priced per minute, commercial use on paid plans |
Text-to-speech and voice
1. ElevenLabs — best all-round audio platform
ElevenLabs is still the reference point for AI voice, and its real advantage in 2026 is breadth. One shared credit pool covers text-to-speech, speech-to-text, voice changing, noise isolation, sound effects, music, dubbing, long-form Studio projects and conversational agents.
- Plans: Free (10,000 credits), Starter $6, Creator $22, Pro $99, Scale $299, Business $990. Annual billing gives two months free.
- Cost: about $0.20 per minute of speech on Starter, falling to about $0.14 on Pro billed annually. API text-to-speech is $0.10 per 1,000 characters on v3, or $0.05 on its faster Flash models.
- Standout detail: unused credits on paid plans roll over for up to two months. Almost nobody else in this list does that, and it matters if your output is uneven.
- Watch out for: it is the most expensive per minute of the voice tools here. You are paying for the breadth and the ecosystem.
Best for: creators and teams who need several kinds of audio and would rather have one bill, and anyone whose monthly volume swings.
2. Fish Audio — best value for voice
Fish Audio is the platform people switch to when ElevenLabs gets expensive, and on English speech the savings are substantial.
- Plans: Free (8,000 credits), Plus $15 or $5.50 a month annually, Pro $100 or $37.50 annually, Max $999 or $749 annually.
- Cost: about $0.028 per minute on Plus billed annually — roughly six times cheaper than ElevenLabs. Its API is pay-as-you-go at $15 per million UTF-8 bytes, about $0.02 per minute of English speech.
- Standout detail: enhanced voice cloning is included even on the free plan, and Voice Design creates new voices from a description.
- Watch out for: credits expire monthly, the Max plan is barely cheaper per minute than ElevenLabs, and byte-based billing makes non-Latin scripts roughly three times more expensive per character than English.
Best for: high-volume English narration and voice-overs, especially on an annual plan. We break this matchup down in detail in ElevenLabs vs Fish Audio.
3. Hume — best for expressive voices and cloning
Hume's Octave text-to-speech models are built around emotional expressiveness — voices that sound like they mean what they are saying — and its pricing is unusually generous on cloning.
- Plans: Free (10,000 characters, about 10 minutes), Starter $3, Creator $14 (half price in the first month), Pro $70, Scale $200, Business $500.
- Cost: about $0.10 per minute on Starter and Creator, falling to about $0.05 on Business. Usage beyond your allowance is billed from $0.15 down to $0.05 per 1,000 characters depending on plan.
- Standout detail: unlimited voice cloning on every plan, including free — create as many voices as you like and use them. Most competitors ration cloned voices by tier.
- Watch out for: it is a voice specialist, with no music or sound effects.
Best for: characters, audiobooks, games and anything where emotional delivery matters more than raw cost — and anyone who needs many custom voices on a small budget.
4. Cartesia — best for real-time voice agents
Cartesia builds for live conversation. Its Sonic models are designed for voice agents that have to answer quickly enough to feel natural, and its plans bundle prepaid agent usage alongside speech credits.
- Plans: Free (20,000 credits), Pro $5, Startup $49, Scale $299, plus custom Enterprise.
- Cost: Pro covers about 133 minutes of speech on its current Sonic model for $5 — around $0.04 per minute, falling below $0.03 on Startup and Scale. That is the lowest subscription price per minute in this guide.
- Standout detail: a commercial licence and instant voice cloning from the $5 Pro plan, with professional cloning from Startup.
- Watch out for: it is optimised for agents and developers rather than for long-form creative production, so compare output quality on narration before committing.
Best for: developers building phone agents, assistants and interactive voice products, where latency and cost per minute dominate.
How the voice tools compare on price
| Tool | Cheapest paid plan | Approx. minutes | Cost per minute | Unused credits |
|---|---|---|---|---|
| ElevenLabs | $6/mo | 30 | $0.20 | Roll over 2 months |
| Hume | $3/mo | 30 | $0.10 | Not stated |
| Fish Audio | $15/mo ($5.50 annual) | 200 | $0.075 ($0.028 annual) | Expire monthly |
| Cartesia | $5/mo | 133 | $0.038 | Not stated |
Treat these as the shape of the market rather than exact rates: each vendor estimates minutes with its own assumptions about characters per minute and model choice. The pattern is still clear — the specialists undercut the all-rounder heavily on pure speech, and the all-rounder earns its premium only if you use what else it offers.
Music generation
5. Suno — best for complete songs
Suno remains the fastest way to go from a text prompt to a finished song with vocals, lyrics and full production. For most people who want "a song about this, in this style", it is the default.
- Plans: Free (50 credits a day, no commercial rights), Pro $8 a month (2,500 credits), Premier $24 a month (10,000 credits).
- Standout detail: Premier lets you record and create with your own voice, and Pro adds custom models.
- Watch out for: downloads are capped separately from credits — 20 song downloads a month on Pro and 60 on Premier. You can generate far more than you can take away, so the download allowance, not the credits, is usually your real limit.
That download cap produces a neat bit of arithmetic: both paid plans work out at $0.40 per downloaded song. Premier's extra money buys room to iterate and your own voice, not cheaper finished tracks. Commercial use requires a paid plan.
Best for: creators who want finished, vocal songs quickly — social content, jingles, demos and personal projects.
6. Udio — best for detailed music control
Udio appeals to people who want to shape a track rather than accept one. It offers longer two-minute generations on paid plans and more hands-on control over style and structure.
- Plans: Free (10 credits a day, 100 a month), Standard $10 or $8 annually (2,400 credits), Pro $30 or $24 annually (6,000 credits).
- Standout detail: paid plans lift the free tier's limits and let you run six or ten generations at once.
- Watch out for: credits do not roll over, and Pro gives fewer credits per dollar than Standard — 200 per dollar against 240 on monthly billing. You upgrade for features and concurrency, not for cheaper credits. Check current export and sharing terms before you build a release plan around it, as they have changed over time.
Best for: musicians and producers who want to iterate on structure and sound rather than take the first result.
7. ElevenLabs Music — best for commercially cautious music
ElevenLabs' music model generates studio-quality tracks from natural-language prompts, up to five minutes long. It has been widely reported as trained on licensed music from label and publishing partners — a meaningful distinction for brands and agencies wary of the copyright questions around AI music.
- Cost: $0.15 per minute through the API, or 900 credits per minute from an ElevenLabs plan.
- Standout detail: commercial use from the $6 Starter plan, drawn from the same credit pool as your voice work.
- Watch out for: it is less of a songwriting playground than Suno. Choose it for licensing peace of mind and integration, not for novelty.
Best for: background music, ads, and brand work where the provenance of the training data matters to legal or procurement.
Sound effects
8. ElevenLabs Sound Effects — best for on-demand effects
Describe a sound — "heavy wooden door creaking open in a stone hallway" — and get a royalty-free effect back. For video editors and game developers it replaces a lot of library searching.
- Cost: $0.12 per minute through the API, or 200 credits per generation from an ElevenLabs plan.
- Standout detail: royalty-free output in 44.1 kHz MP3 or 48 kHz WAV.
- Watch out for: for iconic, recognisable sounds a curated library is still faster. Generation shines for specific, unusual or layered effects no library has.
Best for: video editors, podcasters and game developers who need custom effects without licensing headaches.
How to choose
Four questions settle almost every choice in this category.
- How many kinds of audio do you need? One kind: pick the specialist. Three or more: ElevenLabs' breadth usually beats juggling subscriptions.
- Is it live or produced? Real-time agents need low latency, which points to Cartesia or ElevenLabs Flash. Produced narration can use the slower, richer models.
- Is your volume steady or spiky? Spiky usage favours rolling credits, which ElevenLabs offers. Steady usage lets you take the cheapest per-minute plan.
- Will it be published commercially? Check rights on the exact plan. Suno's and ElevenLabs' free plans are non-commercial, and cloning rules apply everywhere.
Things that catch people out
- Credits rarely roll over. Of the tools here, only ElevenLabs states that unused credits carry forward. Size your plan for an ordinary month.
- Minute estimates are not comparable. Each vendor quotes minutes using different assumptions. Test the same script on each before trusting a comparison.
- Language changes the price. Byte-based billing charges roughly three times more per character for scripts like Hindi, Chinese and Japanese.
- Music downloads can be the real limit. On Suno, credits let you generate far more than your monthly download allowance lets you keep.
- Voice cloning needs consent. Every platform requires rights to the voice you clone. A commercial plan covers your output, not someone else's voice.
A note on these figures
AI audio pricing changes constantly, and several vendors run introductory offers that alter the first-month price. Everything here reflects each vendor's live pricing page as of 18–19 September 2026. Per-minute figures divide plan prices by each vendor's own minute estimates and are indicative rather than measured like-for-like. Always confirm current pricing and commercial terms before committing.
