Audio to Text Converter Formats: MP3, WAV, M4A, FLAC and More

Almost every audio to text converter accepts MP3, WAV and M4A, most also take FLAC, OGG and WebM, and some still reject WMA or oversized files. The format you upload matters less for accuracy than the recording quality inside it, but the wrong choice can cost you a failed upload, a huge file or a lost hour re-encoding. This guide explains what each common format is good for, how bitrate and sample rate relate to transcription accuracy, how to convert files with ffmpeg in one line, and what size and duration limits to expect from typical tools.

The formats you will actually meet

Audio formats are really two things: a container (the file wrapper) and a codec (how the sound is compressed). Most of the time you only need to know whether the compression is lossy or lossless and whether your tool accepts the file.

MP3

The universal format. Lossy, supported everywhere, and perfectly adequate for speech at 96–192 kbps. Older recorders and some voice memo apps produce MP3 at 32–64 kbps, which is where consonants start to blur and transcription accuracy drops. If you have a choice, do not go below 96 kbps for speech.

M4A / AAC

Apple's default for Voice Memos and most iPhone and Mac recordings, and the audio codec inside most MP4 videos. AAC compresses more efficiently than MP3, so a 96 kbps M4A sounds roughly like a 128 kbps MP3. Every mainstream transcription tool accepts M4A. The one trap: some M4A files from Apple devices use the ALAC lossless codec instead of AAC, which produces much larger files and is occasionally rejected by stricter uploaders.

WAV

Uncompressed PCM audio. Nothing is lost, which is why professional recorders, Audacity and many conference systems default to it. The price is size: about 10 MB per minute at CD quality in stereo, 1.9 MB per minute at 16 kHz mono. WAV is ideal for recording and archiving, and a poor choice for uploading long sessions over a slow connection.

FLAC

Lossless like WAV but compressed, typically to 50–60 % of the WAV size. Supported by most transcription services and by every open-source speech engine. If you want to keep an archive master without the WAV bulk, FLAC is the answer.

OGG / Opus

Opus is the codec used by WhatsApp voice messages, Discord, Telegram and most WebRTC-based meeting tools, usually inside an OGG or WebM container. It is remarkably good at low bitrates; a 24–32 kbps Opus voice note is still clearly intelligible to a speech model. Support is now broad, but a few older converters still ask for an MP3, in which case a quick ffmpeg conversion fixes it.

WMA

Microsoft's older format, still produced by some legacy dictation devices and Windows recorders. Support among online transcription tools is patchy. Convert WMA to MP3 or FLAC before uploading; it takes seconds and avoids an error message.

WebM

A container rather than a codec, produced by browser-based recorders, Chrome's MediaRecorder API, Google Meet and many screen recorders. It typically holds Opus audio and VP8/VP9 video. Most modern tools accept WebM directly, including Audiotexta, which also handles the video variants. If one does not, extract the audio to OGG or M4A.

Video containers (MP4, MOV, MKV)

Not audio formats, but the files you most often start with. Any serious transcription tool extracts the audio track itself. The main reason to convert is size: a one-hour 1080p screen recording can be 2–4 GB, while its audio track is under 60 MB.

Format comparison table

Sizes are rough and depend on bitrate and channel count, but the orders of magnitude are what matter when you are deciding what to upload.

Bitrate, sample rate and what they do to accuracy

Speech recognition models are trained on audio resampled to 16 kHz mono, and most services resample whatever you upload to that internally. This leads to a few practical conclusions:

• Sample rate above 16 kHz adds nothing for transcription. 44.1 kHz and 48 kHz recordings are fine; they just take more space.
• Sample rate below 16 kHz removes information. The classic case is 8 kHz telephone audio, where "f", "s" and "th" sounds become hard to distinguish. Expect visibly more errors on call recordings, regardless of tool.
• Bitrate matters for lossy formats. At 128 kbps and above, MP3 and AAC are effectively transparent for speech. Between 64 and 96 kbps they are still fine. Below 48 kbps, MP3 in particular starts to smear consonants; Opus holds up noticeably better at the same bitrate.
• Stereo is unnecessary unless each channel holds a different speaker, which some call-recording systems do deliberately. Otherwise convert to mono to halve the size.
• Re-encoding compounds losses. A WhatsApp voice note (Opus, ~24 kbps) exported to MP3 at 64 kbps and then uploaded has been through two lossy stages. Upload the original whenever possible.
• Converting lossy to lossless does not help. An MP3 turned into a WAV is a big MP3. Only convert when a tool requires it.

In short: for accuracy, the microphone and the room matter far more than the container. Our transcription accuracy tips cover those; this article is about getting the file through the door without losing anything on the way.

Converting with ffmpeg

ffmpeg is the free, open-source tool that nearly every converter, including most transcription services, uses under the hood. It runs on Windows, macOS and Linux, and the official documentation is at https://ffmpeg.org/documentation.html. One command handles the most common job, turning any input (a WMA file, a 3 GB video, a stereo WAV) into a compact mono MP3 that any tool will accept:

ffmpeg -i input.mp4 -vn -ac 1 -ar 16000 -b:a 96k output.mp3

What each flag does:

• -i input.mp4 the source file; the extension can be anything ffmpeg understands, which is practically everything.
• -vn drop the video track.
• -ac 1 downmix to mono.
• -ar 16000 resample to 16 kHz, the rate speech models use anyway.
• -b:a 96k encode the audio at 96 kbps.
• output.mp3 the output; change the extension to .flac, .m4a or .ogg and ffmpeg picks a matching codec automatically.

Swap -b:a 96k for -c:a flac (and .flac as the extension) if you want a lossless result, or use -c:a libopus -b:a 32k with .ogg for the smallest possible upload. If you only need to split a very long file into parts, -ss (start time) and -t (duration) do that without re-encoding when you add -c copy.

Graphical alternatives exist if you prefer not to use a terminal: Audacity exports to MP3, FLAC, OGG and WAV; VLC has a convert dialog; and HandBrake extracts audio from video. All of them are front ends for the same ideas.

File size and duration limits in typical tools

Limits vary widely and change often, so treat this as a map rather than a rulebook:

• Web-based transcription apps commonly cap single uploads somewhere between a few hundred megabytes and a few gigabytes, and recordings at several hours. Free plans are usually stricter on both duration per file and total minutes per month; paid plans raise them. Audiotexta's current limits per plan are listed on the pricing page, and the free tier is enough to test a typical lecture or meeting.
• Speech APIs often impose a per-request limit (tens of megabytes is common) and prefer long files to be either compressed to Opus or uploaded to cloud storage first. If you are integrating directly, read the provider's documentation before designing around a single giant request.
• Local tools such as Whisper have no upload limit but process at a speed determined by your hardware; a multi-hour file on a laptop CPU can take longer than the recording itself.
• Meeting platforms (Zoom, Teams, Meet) store recordings for a limited time and sometimes split long meetings into multiple files. Download promptly.

When a file is over the limit, you have three options in order of preference: convert to a compact format with the ffmpeg command above (a 2 GB video becomes a 40 MB MP3), split it into chapters with -ss and -t, or, for very long archives, use a tool whose paid plan is built for multi-hour files rather than fighting the limit. For video specifically, transcribing video to text walks through extracting and uploading the audio track.

Which format should you choose?

A short decision guide:

• You are about to record and have a choice: record WAV or FLAC at 44.1 or 48 kHz if space is no concern, otherwise M4A/AAC or MP3 at 128 kbps or more. Set the recorder to mono unless you are separating speakers by channel.
• You already have a file and the tool accepts it: upload it as is. Do not convert.
• The file is rejected or too large: run the ffmpeg command above and upload the MP3 (or OGG/Opus for the smallest result).
• The source is a voice message or meeting recording in Opus/WebM: upload directly if supported; convert only if you must.
• You need an archive copy for later re-transcription with better models: keep FLAC.
• You need subtitles rather than a document: the input format does not change; the output will be SRT or VTT, as explained in our how to transcribe audio to text guide.

Checklist before you upload

• Original file, not a re-exported copy
• Format accepted by the tool (convert WMA and odd containers first)
• Bitrate at least 64 kbps for lossy speech, sample rate at least 16 kHz
• Mono unless channels carry separate speakers
• Under the tool's size and duration limit, or compressed and split
• Audio only when the video is not needed
• File named with date and topic so the transcript is findable later

Follow those and the format question effectively disappears. The remaining accuracy differences come from the recording itself and from how you set up the transcription job: language, speaker labels, vocabulary and post-editing, all of which you can test with a single upload to Audiotexta in any of the formats above.