How memozero works: from an audio file to a checked transcript

Karsten B. Kreh & Paul K. Jäckel Karsten B. Kreh & Paul K. Jäckel

This article gives you an overview of your app, from the audio file to the transcript. The transcription runs entirely on your device, without your recording or the transcript leaving the device. Every step has a further, detailed article, and in some cases also an explainer video with a demonstration.

A transcription in five steps

If you want to create a transcription and review it against the audio, proceed as follows after installing and licensing the app:

  1. Submit a file. Click New Transcription and select a file, or drag it into the drop area. You find the details in Start a transcription.
  2. Choose a model. Under AI Model you choose between Whisper 3 Turbo – German, Whisper 3 – Multilingual and Whisper 3 Turbo – Multilingual. Which models your device offers is explained in Models and computing power.
  3. Set the speakers. Under Speaker detection you enter the number of speakers, from 1 to 10. How you name and colour speakers afterwards is described in Naming and colouring speakers.
  4. Listen back and correct. After the processing, the editor opens. There you check the text against the audio and change single words, see Correcting transcripts, Playback and keyboard shortcuts and Search and replace.
  5. Export. With Export in the footer bar you can output the transcript as a Word file, for example, see Export formats.

Between step 3 and step 4, the app works on its own. It shows a list of the processing steps, among them Preparing Audio File, Transcribing with Whisper AI, Word Alignment and Speaker Diarization. Keep memozero open until the transcription is finished. Closing the app stops the running process.

What memozero brings along and what it does not

There is no language selection for the recording. The recognition determines the language itself. You do, however, have to select a German or a multilingual model in advance. For a German conversation you take a German model, for several languages or a foreign language you take a multilingual one.

memozero accepts the common audio formats as well as video files in the formats .mp4 and .mpeg. You find the full list in Start a transcription.

Why local

Local processing of your recordings is, besides the editor, a central part of memozero. It protects your data and saves your resources. The concrete benefits are three:

  • Privacy. Audio and text remain on your device. There is no upload, no share link and no web view of the transcript.
  • No quota. There is no minute limit and no upper limit for the number of transcriptions. You can transcribe when you want and as much as you want.
  • No upload and waiting time. A large file does not have to be uploaded to the internet first. Corresponding privacy and confidentiality declarations thereby become unnecessary.

The other side: the computing time depends decisively on the performance of your device. A device with a supported graphics card or with an Apple silicon chip works much faster than a pure CPU device. With long recordings, the processing can take a correspondingly long time; powerful devices sometimes finish in a few minutes. Which hardware delivers what is described in System requirements and installation.

What memozero calculates for you

The recording gives you more than a simple word text. memozero enables you to use:

  • Word-level timestamps. Every word carries a start time and an end time. The player, the jump to a position and the time entries in the export all build on this. Further analysis programs can build on this. More about it in Playback and keyboard shortcuts.
  • Speaker separation. The text is divided into sections per speaker. In the editor you can assign each speaker a name, a colour or a writing style with one command, for example all contributions of one speaker in bold, see Naming and colouring speakers.
  • Confidence values. Every word carries a value for how sure the recognition by the AI was.

Why memozero also offers an editor

Every transcription, whether automatic or produced by a person, contains errors. If the transcript is to serve as evidence, it must be factually correct. Listen back to it at least once before you process it further or pass it on.

Listening back costs time and concentration, especially with recordings of lower quality, for example when people talk over each other, speak quietly or unclearly, or when background noise occurs. These are circumstances that you do not always control yourself.

That is why it was important to us to support you during the review with targeted functions: the integrated audio player, which lets you jump to individual words and shows where you are in the text, the renaming of speakers including writing style, the dictionary, the search for keywords, the comment function and further shortcuts, see Correcting transcripts. These functions save you time and concentration when you work through the transcript.

How you can process the transcript further

With the export function you can export your transcripts in different formats and with different details:

Word Document (.docx), Text Document (.txt), Markdown (.md) and JSON Data (.json).

Three checkboxes control whether your comments, the word-level timestamps and the processing details go along. You find all the details in Export formats.

Where to go next