Why We Need Speech to Evaluate Speech Translation
This paper highlights a critical gap in evaluating speech translation quality for speech-specific information, which is important for researchers developing speech translation systems that aim to preserve paralinguistic features.
Current evaluation metrics for speech translation fail to assess speech-specific phenomena like gender agreement and prosody. The authors show that both text- and speech-based metrics perform poorly, and even a specially trained model (SpeechCOMET) and a SpeechLLM judge do not consistently capture these phenomena, identifying three root causes.
Speech translation models are increasingly capable of preserving speech-specific information (e.g., speaker gender, prosody, and emphasis), yet evaluation metrics remain blind to such phenomena. We meta-evaluate both text- and speech-based quality estimation metrics on two contrastive datasets targeting gender agreement and prosody, and find that both fall short, even when given direct access to the speech signal. We then train SpeechCOMET, a family of quality estimation models with speech encoders, and evaluate a state-of-the-art SpeechLLM as a judge. Both match or exceed text-based COMET on standard quality estimation, but neither consistently assesses speech-specific phenomena. We identify three causes: (1) speech-specific features are not reliably preserved in current encoders, (2) models tend to ignore the speech source signal, and (3) quality estimation training data contains too few relevant examples. We release all models and code, and argue that progress requires dedicated speech-specific training data and models that genuinely condition on speech.