ElevenLabs Launches v4 Speech Model with Enhanced Expression Control and 90-Language Support

New Speech Model Capabilities

ElevenLabs introduced its v4 and v4 Turbo speech models, building on the v3 model released last year. The new architecture allows for improved control over voice expression and faster voice cloning, requiring only 10 seconds of audio.

The v4 model handles voice identity more consistently over longer text passages and adjusts expressions based on text context. It expands on inline tags for expression definition, allowing users to stack multiple tags for sequential application.

Expanded Language Support

The new models increase language support from 70 to over 90 languages. ElevenLabs noted significant quality improvements for Japanese, Brazilian Portuguese, Mandarin, and Cantonese.

This expansion caters to a broader global user base and enhances the model's utility in diverse linguistic environments.

Enterprise Focus and Performance

ElevenLabs' enterprise calling business accounts for over 55% of its revenue. The v4 model is designed for voice agents, offering lower latency for more fluid conversations.

The model can generate audio as soon as the underlying Large Language Model (LLM) produces responses. It also manages confrontations, escalations, and holds differently to improve issue resolution in agent interactions.

Competitive Landscape and Company Growth

The speech model market is competitive, with companies like Cartesia, Deepgram, Fish Audio, Boson, and WellSaid Labs, alongside Google and OpenAI, developing expressive voice models.

ElevenLabs raised $500 million earlier this year, valuing the company at $11 billion. Its annualized revenue run rate has grown from approximately $330 million to over $600 million, and its headcount has exceeded 800 employees across various markets.