What is scribe?

Scribe is a tool that takes any audio or video file and turns the spoken words into text. What sets it apart from basic transcription is that it can also tell who is speaking when, producing a transcript that reads like a dialogue with each speaker labeled. So instead of getting one wall of text from a recorded meeting, you get a back-and-forth conversation broken out by person. You feed it a file like an MP3 podcast, an MP4 video recording, or a WAV audio clip, and it handles the rest. Behind the scenes, it converts whatever you give it into a standard audio format, then runs speech-to-text to capture the words. The "who said what" part, called diarization, figures out when different people are talking and assigns labels like SPEAKER_00 or SPEAKER_01 to each segment. You get a plain text file as output, or you can request a JSON file that includes precise timestamps for when each person started and stopped talking. This is aimed at people who need accurate, speaker-labeled transcripts of conversations but want the processing to happen locally on their own machine rather than uploading sensitive recordings to a cloud service. That makes it a good fit for journalists transcribing interviews, teams processing internal strategy meetings, podcasters wanting written records of episodes, or researchers working with private interviews. There is a notable tradeoff in how it is built: the tool requires macOS running on Apple Silicon hardware. It leans on Apple's CoreML framework to speed up the speaker-identifying processing, so it will not run on Windows or Intel-based Macs. The required AI models download automatically the first time you use the tool, and no account or API key is needed. You can also skip the speaker-identifying step entirely if you just need a straight transcript without the labels, which is faster if you only care about the words themselves.