ElevenLabs

ElevenLabs is the voice model most other tools quietly run on. Text to speech, voice cloning, dubbing and conversational agents, with output that stopped sounding synthetic somewhere around 2023.

What it does well

Prosody. Most text-to-speech gets words right and emphasis wrong; ElevenLabs reads a sentence like someone who understood it. Voice cloning needs a minute of clean audio to produce something usable, and the dubbing tool keeps the original speaker's voice across languages rather than swapping in a stranger.

The API is properly documented and fast enough for real-time agent use, which is why so many products are built on top of it.

What it costs

Free gets you a small monthly character allowance with attribution required. Paid tiers start around $5 a month and scale by characters, with commercial rights and voice cloning unlocking on the lower paid plans. Heavy dubbing and API use gets expensive quickly, and characters are consumed by regeneration, so budget for the takes you throw away.

Where it falls short

Long-form consistency drifts. Generate a full audiobook chapter in pieces and you can hear the seams. Non-English output is strong but noticeably weaker on tonal languages. And the character-based pricing punishes iteration, which is exactly what you do while finding the right read.

Who it's for

Video creators who need voiceover without recording, developers building voice into a product, and anyone localising existing content. If you only need occasional narration, the free tier plus a competitor's free tier will cover you. If voice is core to what you ship, this is the one to buy.