Can you transcribe MP3, M4A, OGG and other audio files?

Yes. MP3 and M4A are on OpenAI's list of accepted formats, and OGG may work in open-source Whisper. The format that matters is the one your chosen tool accepts, not the one your recorder saves. OpenAI's transcription API accepts seven formats, including MP3, M4A, WAV and WebM, but not OGG, so an OGG file must be converted before you send it there. Open-source Whisper passes files to ffmpeg, which reads more formats, so OGG may work there too. Check the tool's list, convert when it is missing, and test a short clip before you transcribe a long recording. The guide to transcribing audio to text explains the rest of the process.

MP3 and M4A are on OpenAI's list of accepted formats

OpenAI's speech-to-text guide says supported input formats are mp3, mp4, mpeg, mpga, m4a, wav and webm ¹. MP3 and M4A are on that list, so those files go straight to the service. Many phone and voice recorders save in one of these two formats, so the first file you try is often accepted without changes.

The table shows what each route takes. The open-source Whisper column depends on how Whisper reads audio, which is explained below.

OGG is the format to convert first

OGG is not on OpenAI's list, so convert an OGG file before you send it to that API. Google's Speech-to-Text API does name OGG_OPUS as an audio encoding, and for multichannel recognition it accepts from 1 to 8 channels for that encoding ². That shows Google supports the encoding. It does not show that every tool does, so the tool's own list is what counts.

Open-source Whisper reads files through ffmpeg, so OGG may also work

Whisper's audio loader starts the ffmpeg command-line tool as a subprocess to decode each file, and it requires ffmpeg in the PATH ³. The loader does not list accepted formats, so the practical limit is whatever ffmpeg can decode. In our reading of the code, that includes OGG. We have not tested an OGG file in this way, so run a short clip before you transcribe a long recording.

The Whisper README also requires ffmpeg to be installed on your system ⁴. Install it first, and check that it runs from the command line.

whisper.cpp needs 16-bit WAV, so convert other files first

The whisper.cpp command-line example runs only with 16-bit WAV files, so audio in other formats must be converted before transcription ⁵. The whisper.cpp README gives an ffmpeg example for the conversion. A command for an OGG file looks like this:

ffmpeg -i input.ogg -ar 16000 -ac 1 -c:a pcm_s16le output.wav

Whisper resamples all audio to 16,000 Hz on its own ⁶, so you do not need to match that rate yourself. The conversion is about making the file readable by the tool, not about raising accuracy. Keep the original file as well.

Size limits matter more than format for long recordings

OpenAI's Transcriptions API accepts files up to 25 MB. Larger recordings must be compressed or split into chunks ⁷. A long recording in a high-quality format can pass that limit, so check the file size before you upload it.

Compressing to a lossy format brings back the accuracy risk described in the question on low-quality audio, so splitting is usually the better choice. Cut at a pause between sentences, and check the first words after each cut against the audio, so no sentence is lost at the join.

Steps to check a format, convert it and test the result

1. Check the tool's list. Look in the tool's own documentation for the accepted formats and size limit, since they differ by service.
2. Convert if the format is missing. Use ffmpeg to write a WAV file, at 16-bit for whisper.cpp, and keep the original.
3. Transcribe a short sample first. Use ten seconds of clear speech and compare the text with the audio.
4. Split a long file if needed. Stay under the size limit and cut between sentences.
5. Proofread the full transcript. Use the steps in the guide to transcribing audio for names, numbers and quiet parts.

If your recording is hard to hear, the question on low-quality audio explains how to improve it before you convert. For a free route that accepts your format, see how to transcribe audio for free. For the privacy side, see how to transcribe audio offline.