MP4 to Transcript
Research tool
MP4 to Transcript turns a recording of a meeting, interview or lecture into a transcript laid out like Zoom's, with each voice given an anonymous number: Speaker 1, Speaker 2, and so on, in the order the voices first speak. The recording stays on your computer. The page runs three speech models inside the browser tab, from files on this site, and sends the recording to no server.
Whisper base (English) writes down what is said. Pyannote's segmentation model finds where voices speak and change, and a WeSpeaker model makes a voiceprint of each stretch of one voice; voiceprints that are alike are grouped into one speaker. The numbers come from how voices sound, so two people with similar voices may share a number, and one person may get two. On one recorded four-person meeting from the AMI corpus, the page found four speakers and put 95% of the transcript's time under the right one.
Using the tool
Choose a recording: an MP4 or MOV video, or M4A, MP3, WAV or WebM sound. If you know how many people speak, choose the number; otherwise leave Find out pressed. Select Make the transcript, then save it as VTT, in the layout of a Zoom cloud recording's transcript, or as text, in the layout of a transcript saved from a Zoom meeting. The first use downloads about 112 MB of models. A 17.5-minute meeting took about five minutes on a recent laptop.