Coda v2

Coda v2 is a flagship text-to-speech model developed by Rime, engineered specifically for real-time conversational agents and customer experience environments. The model combines a language model backbone with a dedicated speech inference engine trained on conversational data, achieving sub-100ms latency on GPU engines. Coda v2 provides extensive multilingual support across multiple languages and hundreds of distinct voices.

Key capabilities of Coda v2 include word-level timestamps for precise text-audio alignment, built-in interruption handling, and streaming text input support that enables audio generation as an LLM writes its response. Version 2 introduces advanced custom pronunciation control, allowing developers to supply custom lexicons using X-SAMPA phonemes or record voice samples to define word pronunciation across multiple languages without model retraining.