How to find word error rate of spoken sentence for regression based model?

Word Error Rate (WER) is a widely used metric for evaluating Automatic Speech Recognition (ASR). To calculate WER for a visual speech synthesis (VSS) system, a reference word transcription and a hypothesis word transcription will be needed, and then standard word error rate alignment can be performed to obtain the WER. These word transcriptions can be obtained in various ways. For example, the reference word transcriptions might come from the visual dataset labels. The hypothesis word transcription might come from the VSS system itself (if the VSS system has an intermediate representation in words), or from running ASR on the synthesized speech. It is important to note that while WER is a widely-used metric, it does not capture all aspects of visual speech synthesis quality. Other evaluation metrics, such as perceptual evaluation of speech quality (PESQ) or subjective user studies, could be conducted to assess the system's performance from different perspectives, including audio-visual synchronization, intelligibility, overall usefulness of the synthesized speech, and naturalness.