Product

Local AI Audio Transcription

Batch transcription of long form video and audio files using open models, local processing and low cost infrastructure.

Problem

Needed to transcribe videos exceeding two hours on a recurring basis. Using paid services would increase costs and require uploading large files, making the processing workflow more challenging.

My contribution

Developed a solution that splits content into short audio segments and uses faster-whisper to generate transcripts. Processing can run on CPU or be accelerated with NVIDIA GPUs via CUDA, depending on available resources.

Outcomes and learnings

Enabled low cost batch transcription of long form content by leveraging available hardware and keeping processing local. The project demonstrated how open models can address a recurring need without relying on paid transcription services.