ElevenLabs Turbo v2.5 is a text-to-speech (TTS) model designed for high-speed, low-latency audio generation. It is engineered to facilitate real-time conversational AI applications by providing rapid response times while maintaining high vocal naturalness. The model supports 32 languages, including English, Hindi, French, Spanish, and Mandarin, and introduced support for Vietnamese, Hungarian, and Norwegian upon its release.
Compared to the previous Multilingual v2 model, Turbo v2.5 is approximately three times faster for non-English languages and 25% faster for English. It is optimized for high-throughput environments and can handle requests up to 40,000 characters, which equates to roughly 40 minutes of audio. The model architecture balances synthesis quality with speed, making it suitable for interactive use cases such as virtual assistants, gaming, and real-time translation.
In addition to performance gains, the model incorporates fine-tuned versions of default voices to enhance clarity and prosody. Users can adjust the output through control parameters such as stability and similarity, which allow for customization of the emotional range and adherence to the original speaker's identity.