Voice Memo to Text From Your Phone to a Transcript

Which file your phone actually saves

The Voice Memos app on iPhone saves every recording as an M4A file, and that is the format you get when you tap Save to Files or AirDrop a memo to a Mac. Android is less uniform. Samsung Voice Recorder writes M4A, the Google Recorder app exports M4A as well, and older or lighter recorder apps still produce 3GP or AMR files from the days when storage was scarce. All of these go into the upload box above without any conversion step.

The extension is not something you need to check before uploading. If the recorder on your phone saved it, the chances are high it is on the list of accepted formats, and if a file is refused you will see a message straight away rather than after a long upload.

What a voice memo transcript is good for

Most voice memos are made because typing was not an option at the time. You dictate an idea on a walk, record the main points after a client meeting in the car park, capture the five minutes of a lecture where the examiner hinted at the exam, or hold the phone up during an interview because the proper recorder stayed at home. The recording is quick to make and slow to use later, because finding one sentence in a twenty-minute memo means listening to the whole thing.

A transcript turns that recording into something you can search, skim and paste. The timestamps let you jump back to the moment a particular point was made, and the text can go straight into a document, a to-do list or an email. For journalists working from phone recordings, the transcript is also the fastest way to pull an accurate quote.

Recording quality from a phone

Phone microphones are good at picking up the person holding the phone and much worse at a voice across the room. If you are recording someone else, hold the phone closer to them than to you, or put it on the table between you rather than in your hand. Wind is the enemy of outdoor memos, so a hand cupped around the bottom of the phone makes a real difference on a walk.

The model detects the language automatically and handles more than fifty of them, so a memo that switches between two languages comes out correctly. If the memo contains two or more people, each speaker gets a label that stays consistent through the file.

The memo stays with us, not with a third party

The speech models run on our own servers, and the recording is not forwarded to a third-party speech API for processing. That matters for memos, because they tend to contain the unpolished version of your thinking, names of clients and colleagues, or the words of someone you interviewed. The same applies to the transcriptions you make here as a guest; if you later create a free account, those transcripts move into it automatically.