This usually happens because the model isn’t tuned for things like emails/names or the audio quality + accents are throwing it off. For better accuracy, you should look at options like Deepgram, AssemblyAI, or OpenAI Whisper they handle entities (emails, names) much better, especially with proper settings like custom vocabulary or post-processing. Quick question: what kind of audio are you feeding it (calls, mic, VoIP), and do you need real-time or just recordings?