EMOVA-ASR-TTS-Eval is a dataset designed for evaluating the ASR and TTS performance of Omni-modal LLMs. It is derived from the test-clean set of the LibriSpeech dataset. This dataset is part of the EMOVA-Datasets collection. We extract the speech units using the EMOVA Speech Tokenizer.