Hume AI has officially released TADA, a new text-to-speech (TTS) system based on a unique Text-Acoustic Dual Alignment tokenization approach. This solution is now available to the public under an open-source license, with models and code accessible for immediate use. TADA targets researchers, developers, and companies building voice-enabled applications, offering both English and multilingual models.
Test voice generation on Hume AI Platform 👀
The product stands out for its speed, producing real-time speech over five times faster than comparable LLM-based TTS systems, while maintaining high accuracy and nearly eliminating content errors. TADA’s one-to-one text-to-audio token alignment allows it to generate long-form speech, with a context window that supports up to 700 seconds of audio, far surpassing existing solutions that struggle with context limitations, memory demands, and hallucinated or missing speech content.
Today we're releasing our first open source TTS model, TADA!
— Hume AI (@hume_ai) March 10, 2026
TADA (Text Audio Dual Alignment) is a speech-language model that generates text and audio in one synchronized stream to reduce token-level hallucinations and improve latency.
This means:
→ Zero content hallucinations… pic.twitter.com/4JMQSghqCz
Hume AI, the company behind TADA, specializes in voice AI research infrastructure, catering to AI-focused organizations and research labs. Its mission is to advance reliable, efficient voice generation technologies. The release of TADA marks a strategic move to drive progress in TTS through open-source collaboration, providing tools for on-device deployment that lower latency and address privacy needs. Early technical evaluations indicate high speaker similarity and naturalness scores, positioning TADA as a strong alternative to both previous versions and competing TTS offerings. Initial responses from developers and AI experts highlight the potential for this architecture to reshape voice synthesis in regulated and resource-constrained environments.
