Evaluation benchmark and framework for testing advanced TTS capabilities.

We present ARECHO, a token-based autoregressive speech assessment framework that models inter-metric dependencies to enable accurate, interpretable, and flexible evaluation across diverse speech tasks.