Audio transcription software: three ways a recording becomes text
Audio transcription software turns recorded speech into written text, and there are three ways to have it done: people who listen and type, a cloud service you upload the file to, or an app that recognises the speech on the computer the file is on. Dalm is a Mac app of the third kind. You bring in an audio or video file up to four hours long, on the Free plan; the speech is recognised on your Mac and the audio is not sent anywhere, while the text is synced to our servers automatically. The original file stays where it is and is not copied.
What people transcribe: interviews, lectures, calls and voice memos
Four common kinds of recording, and how each one starts in Dalm.
- An interview. Two or more people, and you will want their exact words for quotes. Bring it in as a meeting with several voices.
- A lecture or a talk. One voice for most of the recording. Bring it in as a meeting so it gets a summary: a dictation stays plain text and cannot be turned into a meeting later. Anything past four hours has to be split into parts first. For students, recording lectures has its own page.
- A recorded call. A file a call app saved, or one a colleague sent you. It comes in as a meeting. A call you have not had yet needs no file at all: Dalm records calls on the Mac from your microphone and the sound of the call, and no bot joins the meeting.
- A voice memo. A thought recorded on a phone. Once the file is on your Mac, it comes in as a dictation and sits with your other dictations.
How Dalm turns a file into text, step by step
- You say what it is. Dictation, or a meeting with several voices. The same sheet names the family of recognition models the file will go through, taken from your settings.
- The file is read in place. Dalm opens it in the folder it is in, piece by piece, and does not copy it; the temporary working file made on the way is removed right after. If there is no sound in it, the app stops and tells you.
- The speech is recognised on your Mac. The language is detected for you within the family you chose, and while the audio is not sent anywhere, the text is synced to our servers automatically.
- Your dictionary is applied. Once recognition is done, names and terms you have added replace the ways they tend to be misheard.
- You watch it or leave it. A bar counts the minutes of the recording that are done and estimates how many are left. Close the window and the transcript still turns up in your list when it is ready.
Whether any of this needs a network is answered on a page of its own.
People, an upload or an app: what each gives and takes
| People who transcribe | A cloud service you upload to | An app on your Mac, such as Dalm | |
|---|---|---|---|
| Who turns speech into words | A person listens and types | A speech model on the provider’s servers | A speech model on your Mac |
| What you gain | A person decides what was said where the sound is unclear | No software to install; any computer with a browser | No upload step; the file stays in its folder; files up to four hours on the Free plan |
| What it costs you | Someone outside hears the recording, and you wait for their turn | The file has to reach their servers first, and what happens to it there is set by their terms | A Mac with Apple Silicon · macOS 15.0 or later, a model download on first launch, and a transcript that is what the model produced, plus your own corrections |
| What to ask first | Is it people only, or a machine draft that a person checks? | How long may one file be, how large, and how long is it kept? | Does the text leave too? In Dalm it does: it is synced to our servers |
None of the three suits every recording. If you need someone to answer for each word, the first column is the one. If the recording is not private and there is no Mac to hand, the second. If you would rather the recording itself stayed with you and you work on a Mac, the third — knowing that the text still goes to our servers. The same choice with names attached is seven tools, where each is better, and whether a recording should leave your computer at all is the question the guide asks before anything else.
What changes the result, and why no score is published here
Three things you control shape the transcript. The first is the recording itself — what it does to the result is covered in the guide. The second is the family of recognition models: Dalm detects the language, but within the family you set, European languages, which include English, or More languages, a much wider set. The sheet you see when bringing a file in names the family it will use; to switch, change it in Settings under Recognition before the file goes in. The third is the words no model has heard: people’s names, product names, in-house terms. Put them in the dictionary with the ways they get misheard before you bring the file in. The dictionary works for dictations and meetings alike, and a file you bring in becomes one or the other.
You will not find an accuracy percentage or a processing time on this page. A figure goes on this site only with a record of how it was measured — on what recording, on which machine, how many times — and for transcribing files we do not have that record. Describing the result in words instead would be the same claim with the evidence taken out.
Limits of a file you bring in
- Length. From 0.3 seconds to four hours per file; a longer one is turned away with a message saying so.
- Size. How large the file is makes no difference: it is not uploaded, and only the text made from it goes to our servers.
- Sound. A file with no audio track is refused, and of a video only the sound is used — transcribing a video covers what that leaves out. If the model cannot make out a single word, the app says that too.
- File types. Anything macOS reads as audio or video; the detail, and the steps, are under transcribing a file on the Free plan.
- Language. Dalm writes down what was said in the language it was said in; it does not translate.
Free, Pro or Team for transcribing recordings
Bringing in a file is part of the Free plan, and Free has no end date: it is not a trial that turns into a bill. Dalm asks for a Mac with Apple Silicon · macOS 15.0 or later and an account, and while the app is not public, access goes through the waitlist. What the paid plans add is on the pricing page. A team shares one dictionary, templates and tags — how that works for a team. Other jobs the Free plan does, one task at a time, are under free transcription tool for Mac.
Questions
- What is audio transcription software?
- Software that turns recorded speech into written text. It works with a speech recognition model, which runs either on a server the file is uploaded to or on the computer the file is on; services that use people have a person type it instead. Dalm is a Mac app of the second kind: the speech is recognised on your Mac and the audio is not sent anywhere, and the text is synced to our servers.
- Is human transcription better than software?
- They trade different things. With people, someone listens to the recording and decides what was said where it is unclear, and you wait for their turn. With software, a speech model does it, and you check names and unclear passages yourself. In Dalm the speech is recognised on your Mac, the audio is not sent anywhere, and the text is synced to our servers.
- How accurate is Dalm?
- We do not publish an accuracy score: a figure goes on this site only with a record of how it was measured, and for transcribing files we do not have one. What you can change is the recording, the family of recognition models in Settings, and a dictionary of names and terms.
- How long does Dalm take to transcribe a recording?
- We do not publish a time, for the same reason. The file is read where it is rather than uploaded — the audio is not sent anywhere, and only the text goes to our servers — and a bar shows the minutes done and an estimate of those left. You can close the window while it runs.
- Does Dalm translate the recording?
- No. It writes down what was said in the language it was said in, detecting the language within the family of models you chose.
- Does the file leave my Mac?
- No: Dalm reads it from the folder it is in and does not copy it, and the audio is not sent anywhere. The text made from it does leave: it is synced to our servers automatically, and when the app writes a summary or a document, a third-party language model gets that text and never the sound.