AssemblyAI’s Universal-2 Speech-to-Text model outperforms OpenAI’s Whisper models in accuracy, proper noun detection, and reduced hallucination rates, according to a recent report. Universal-2 recorded the lowest Word Error Rate (WER) at 6.68%, a 3% improvement over its predecessor Universal-1, while Whisper models had slightly higher error rates.
The Universal-2 model also had better performance in proper noun recognition and text formatting. However, Whisper large-v3 showed strength in alphanumeric transcription. Despite its strengths, Whisper’s susceptibility to hallucinations makes it less reliable for real-world uses.
