MP3 to text converter: Transcribe your audio file without uploading it

MP3 to text, step by step

1. Sign in and open New transcript.
2. Choose the language spoken in the recording, one of 99.
3. Select File as the source, drop your MP3 on the field and press Continue.
4. On the page that opens, the whole recording is selected and processed in the background. For only a part of it, press Choose a range.
5. Once the model is ready, press Start transcription. It downloads only on first use.
6. Read and correct as the text comes, then export it.

Transcribe only part of a file

If you need twenty minutes of a two-hour recording, you do not have to wait for all of it. Press Choose a range, then set it on the time bar or in the start and end fields:

Three things worth knowing:

• Times in the text count from the start of the file, not from where you began, so a quote keeps its place in the recording.
• You can add another range to the same transcript later.
• The 30 minutes of the free plan go to the part you actually need. If you know the moments in advance, taking timecodes during the recording helps.

Long recordings

In the background the file is not played; your device works as fast as it can. With a graphics card that is much shorter than the recording. Which model to choose depends on your device and how accurate you need it, and choosing a model walks through that.

When a recording is very long, three things come into play:

• Memory limit: Devices with 8 GB of memory or more decode files of up to 120 minutes at once, others up to 45 minutes. Longer files are played in real time in the background, so they take as long as the recording.
• A transcriber device: If your computer is weak, a stronger one of yours or a teammate’s can do the work, also in real time.
• The cloud: To skip the model download, use your own API key with a provider such as Groq or OpenAI, or buy a credit package. The file goes in 60-second pieces.

Is the file uploaded?

Not on the default route. With This device selected, the audio stays in your browser and LiveType’s server holds only text and account details. On the other routes it leaves your device by your choice:

• Transcriber device: It goes straight to that device over an encrypted connection, without touching our server.
• Cloud with your own key: Pieces go straight from your browser to the provider, whose terms apply.
• Cloud with credits: Pieces pass through our server, unsaved, on the way to the provider.

The details are in the privacy notice. For a confidential recording, stay on This device. Other services work by upload; see how UniScribe differs on its comparison page.

Correct early, not at the end

While an hour-long interview transcribes, you can fix a wrong name in the first ten minutes. Your correction never mixes with the new text.

With speaker separation on, name each speaker once. Then export as DOCX, ODT, TXT, MD, or timed SRT and VTT.

To move the text into a document see the MP3 to Word route, and for conventions see how to transcribe a recording.

What MP3 transcription will not do

• It does not translate or summarise: it writes what was said.
• It is not for song lyrics. The engine looks for speech and tries to skip music.
• It needs an account. The free plan has 30 minutes in total, and they do not renew.

If what you have is a stream still running rather than a file, live stream transcription is the better fit.