You should implement the new nvidia parakeet and canary models. They excel in multi lingual speech to text performance and are really fast. So far no API is available in the market. This a nice opportunity for groq! Your current whisper model is really inaccurate in comparison.
Check out the the multi lingual benchmark of Open ASR for proof about accuracy and speed:
https://huggingface.co/spaces/hf-audio/open_asr_leaderboard
https://huggingface.co/nvidia/canary-1b-v2
https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3