Show HN: Sys2txt – transcribe system audio to text in the CLI with a local LLM (Joe-Heffer.github.io)

Sys2txt, a new command-line tool, allows users to transcribe system audio to text leveraging local large language models (LLMs). By utilizing PulseAudio or PipeWire for audio capture and ffmpeg for recording, Sys2txt can operate in two modes: an on-demand recording that transcribes audio once stopped and a live mode that continuously segments and transcribes audio every few seconds. This innovative tool integrates several transcription engines, including faster-whisper for optimal speed on CPU and NVIDIA GPUs, openai-whisper as a baseline reference, and whisper.cpp for those using AMD GPU support via Vulkan. The significance of Sys2txt lies in its ability to offer a seamless, local audio transcription solution without relying on cloud services, which is particularly valuable for privacy-sensitive applications. By automatically selecting the most suitable transcription engine installed on the user's system, it maximizes efficiency while minimizing setup hurdles. This ease of use and flexibility could greatly enhance productivity for developers and researchers in the AI/ML community, enabling them to convert audio data into textual formats readily for further analysis and processing.