MP4 to text

MP4 is a container, not a codec

This distinction explains most of the confusion around MP4 files. MP4 is a wrapper that holds separate streams — typically one video stream and one audio stream, sometimes subtitles as well. The actual compression is done by codecs inside that wrapper, most often H.264 for the picture and AAC for the sound.

That is why two files that both end in .mp4 can behave completely differently in a video editor: they may contain different codecs. For transcription it matters less, because only the audio stream is read, and the audio in an MP4 is almost always AAC, which is well supported.

It also explains why MP4 files are large. You are storing a full video stream alongside the speech you actually want. A one hour meeting recording can easily run to several hundred megabytes, almost all of which is picture that gets discarded during processing.

Where your MP4 probably came from

Phones are the most common source. Both iPhone and Android record video as MP4 or MOV, and the audio quality depends heavily on how far the phone was from the person speaking. A phone propped up across a room captures the room; a phone held close captures the voice.

Meeting applications are the next most common. Recordings from Zoom, Teams and Meet are usually MP4 with a mixed audio track containing every participant. These transcribe well apart from moments where people speak simultaneously, which is unavoidable once the voices are mixed into one track.

Screen recorders produce the cleanest results of all. There is a single speaker on a headset or laptop microphone, no room noise, and no camera distance. If you are recording something specifically in order to transcribe it, a screen recording is the format to choose.

Working with large files

The uploader shows the maximum file size accepted, and long meeting recordings can exceed it. There are two ways around this and one of them is better than the other.

The better option is to reduce the video quality before uploading, using any video tool. Dropping a recording to 720p or lower typically shrinks the file dramatically and has no effect at all on the transcript, because the picture is discarded anyway. This is the rare case where losing quality costs you nothing.

The other option is to extract just the audio and upload that instead. It produces a much smaller file for the same speech. It is more effort, and it makes no difference to accuracy, so it is mainly worth doing if your connection is slow.

Splitting a very long recording into parts also works, and has the side benefit of moving through the queue faster, since shorter files finish sooner.