Feature Request: Add NVIDIA NIM Transcription Backend for Voice Notes

Feature Request: Add NVIDIA NIM Transcription Backend for Voice Notes: a task in LegoFlow-SWE (Harbor dataset). Currently, the voice note transcription system ( transcribe audio ) only supports local Whisper (Hugging Face transformers ) using CPU or CUDA. We need to add an NVIDIA NIM backend that…

The task

Currently, the voice note transcription system (`transcribe_audio`) only supports local Whisper (Hugging Face `transformers`) using CPU or CUDA. We need to add an **NVIDIA NIM** backend that uses NVIDIA’s Riva gRPC API, giving users a cloud‑based alternative that does not require a GPU locally.

Part of Lego-X/LegoFlow-SWE.