Productivity#Text to speech#Voice cloning
ElevenLabs Eleven v4: a voice model that reads the script — speakers, emotions and sound effects annotated
Eleven v4 reads the script: speaker turns, emotions, pauses and sound effects as text annotations; 90+ language cloning and an AI canvas. SaaS.
Product facts
Commercial Suite- Pricing
- Free tier + subscription
- Data checked
- 2026-11-07
Capabilities
- Script-style speech generation (speaker/emotion/pause tags)
- 90+ language voice cloning
- AI canvas for dubbing and sound design
- v4 Turbo low-latency tier
Snapshot figures reflect the check date and may change over time.
The next step for speech synthesis is not “sounds more human” — it is “acts the script.” ElevenLabs’ Eleven v4 turns that into a product shape with an entirely new architecture: the model reads a script the way a voice actor would. It knows who’s speaking, what just happened, and how every line should land. Speaker turns and emotions are annotated directly in the text ([warm], [whispered], [long pause], [crowd applause]), and a multi-speaker dialogue generates in one pass. The official demo is a quiz show — host and two contestants, each with their own voice, with pauses, applause and a gong all driven by the annotations.
Core Purpose
- Script-style generation: speaker turns, emotion cues, pauses and sound-effect events written as text annotations and performed by the model.
- 90+ language cloning: instant and professional cloning across 90+ languages per the official page.
- v4 Turbo tier: a low-latency variant for realtime conversational streams.
- AI canvas: voice, sound effects and music arranged on one canvas — from dubbing to finished piece in a line.
- Pipeline integrations: official integrations for video, audiobooks and podcasts.
Key Scenarios
Content teams producing audiobooks, podcasts, ads and short video; any scene that needs multi-speaker dialogue in one take. Game NPCs and interactive voice are natural users of the script mode too.
Trying it and its limits
The free tier tries v4 (MP3 44.1kHz output to start), subscription billed by credits. Honest limits: closed-source SaaS; voice cloning requires the owner’s consent; emotion annotations steer the performance rather than guarantee it, so long scripts still need human tag-tuning; Chinese support should be validated against the official 90-language list.
One-line verdict
v4 pushes TTS from reader to voice actor — one annotated text, one pass, a finished dialogue. The open-source counterpart on this site is VoxCPM2 (Tsinghua’s 2B model, local GPU deployment): choose open source for data privacy and cost ceilings, choose v4 for out-of-the-box acting ability and 90-language cloning.