Start a Transcription: File, Model, Speakers, Queue
Every transcription starts the same way. You add a file, pick a model, give the number of speakers, and confirm with Continue. After that, your device transcribes the selected audio file with the selected settings. The file does not leave your device in the process; memozero processes it locally. This article walks you through all the steps and shows what happens in the queue and when something fails.
Step by step
If you want to start a new transcription with memozero, proceed as follows:
- Click New Transcription in the sidebar or on the start page.
- The Open File card appears, with the line “Select an audio file to transcribe.”
- Drag your file into the drop zone (it reads “Click to select audio file or drag and drop here”). Alternatively, click the zone and pick the file in your folders.
- Under AI Model, select a suitable model (see below).
- In the Speaker detection section, set the number of speakers with minus and plus. Leave the Auto-Detect speakers switch off if the number is known to you.
- Click Continue. While the job starts, the button reads “Processing…”.
memozero remembers the model, the number of speakers, and the switch for the next time. On the first start, it shows one speaker, and automatic detection is off.
Which files memozero accepts
memozero transcribes audio and video files in these ten formats. The Open File card lists them as well.
- Audio:
.mp3,.wav,.m4a,.flac,.aiff,.aif,.ogg,.opus - Video:
.mp4,.mpeg
For video, only .mp4 and .mpeg work. The app rejects other video formats. In return, you do not have to extract the audio track first. That saves you the detour through a second program, especially with a recorded interview.
memozero checks only the extension, not the content. The app rejects an empty file immediately.
Choose the right model
Under AI Model you can choose between three models for transcription. The basic question is whether a recording was held only or almost exclusively in German, or in another language or in several languages.
| Label in the app | What the app says about it |
|---|---|
| Whisper 3 Turbo – German | “Specially trained for German, most accurate on German recordings, and fast.” |
| Whisper 3 Turbo – Multilingual | “Faster processing for recordings in other languages.” This is useful, for example, for an English-language recording or a recording with an interpreter. |
| Whisper 3 – Multilingual | For recordings in other languages, at the highest accuracy, but with a longer transcription time. For purely German recordings, the German Turbo model is better suited. |
Which models you see depends on your device. On devices without graphics acceleration, the same models also appear with the addition (CPU). If you have a supported graphics card or an Apple Silicon chip, you do not see the CPU rows. A Mac with less than 16 GB of memory gets only a leaner version of the German Turbo model. A model that is not on your device yet carries the addition (Download).
Specify the speakers
The Speaker detection section gives you two controls. They exclude each other.
- Fixed number (default). A counter from 1 to 10. memozero gives the speaker detection exactly this number. If you know the number, enter it. The result then becomes more accurate.
- Auto-Detect speakers. A switch; it is off by default. The software then counts the speakers itself.
If a speaker in a longer recording says something only a few times, we recommend not counting this speaker when the audio quality of the recording is not very good. A rare speaker can then often be assigned more easily in the editor afterwards.
You cannot turn speaker detection off. The switch only changes whether memozero gets the number in advance or counts on its own. There is no range such as “between 2 and 5 speakers”.
On older devices that cannot do speaker detection, both controls are locked. They show “Speaker diarization not supported on this device”. You still get a full transcript. All words then carry an unknown speaker.
What happens during processing
After you click Continue, memozero shows a list of the steps. The first three rows appear only when they are needed.
| Step | What happens |
|---|---|
| Downloading AI Engine | only the first time |
| Downloading Model | when the selected model is missing |
| Waiting in Queue | when another job is running |
| Preparing Audio File | the audio file is prepared |
| Transcribing with Whisper AI | speech becomes text |
| Word Alignment | precise timestamps are calculated |
| Speaker Diarization | the different speakers are identified |
| Finalizing Results | the transcription is prepared for the editor |
Above the list, an overall bar runs with the label “Overall Progress”. During a download, it reads “Download Progress”, and in the queue “Queue Status” with your position. Below the running step, a detail line runs along. After the transcription finishes successfully, the status changes to Review, and the editor opens.
The queue
Only one transcription runs at a time (even on powerful devices). Further jobs wait. There is no upper limit for the waiting jobs. The order is the order of arrival.
You can cancel a running job and a waiting job. To do so, click Cancel Transcription. The app asks “Cancel Transcription?” and adds: “Are you sure? You can restart it from the beginning later.” A cancellation discards the interim results. The file and the entry stay, the status becomes Canceled, and the button then reads Restart. A manual restart puts the job at the end of the queue.
Caution: If you close memozero, you stop the running transcription. The app does not warn you about this. On the next start, memozero puts the job back into the queue automatically and tries again; in total, up to three attempts are possible. The advice from the app is: “Keep memozero open until the transcription is complete.”
Limits
| Limit | Value |
|---|---|
| Number of speakers | 1 to 10 |
| Waiting jobs | unlimited |
| Jobs running at the same time | exactly 1 |
| Length of the recording | unlimited |
| File size, hard limit | 50 GB |
| File size, warning | from 2 GB, from 1 GB with 8 GB of memory |
There is no minute quota and no upper limit for the number of transcriptions. The limits are your hardware and your patience. With a large file, memozero shows this warning: “Large files are supported, but offline processing can take a long time. Keep memozero open until the transcription finishes.”
When something goes wrong
If something goes wrong, you can receive the following messages with the following meaning:
| Case | What you see |
|---|---|
| wrong extension | “Unsupported file format” |
| no audio in the video file | “This file contains no audio track and cannot be transcribed.” |
| damaged or encrypted file | “This audio file could not be read. It may be damaged, encrypted, or in an unsupported format.” |
| little free memory | dialog “Low memory available”, button Start anyway |
| model missing | dialog “Download required” with the size, button Download & Start. The transcription then starts on its own |
| model files damaged | dialog “Model files damaged”, button Download again |
| service restarts | notice “Transcription service is restarting…” |
| other error | “Transcription failed”, button Go back and try again |
Every error message about the audio file carries a support ID. Write this ID down and put it into your support request. With it, support finds your case immediately.
If the progress of the transcription or the speaker detection stays at the same percentage for a long time, give memozero generous time, especially on older devices; for long recordings, leave the program running overnight if you can. If the status does not change after several hours, cancel the transcription, restart your device, and then start the transcription again. If the error appears again, export a support package in Account Settings on the System tab under Diagnostics and send it to us through the contact form. We then aim for a quick response.
Where to go next
- Models and computing power: which model fits your hardware and how much disk space it needs.
- Correcting transcripts: the editor, the confidence, and the way from the status Review to the finished transcript.
- Dictionary: store your own names and technical terms, so that memozero writes them correctly.