AAC or MP3? How Two Lossy Audio Codecs Differ in Practice

Two generations of the same idea

MP3 is short for MPEG-1 Audio Layer III, a codec standardized in the early 1990s. It throws away parts of the sound that a psychoacoustic model predicts you will not hear, such as quiet tones masked by louder ones nearby, and stores the rest compactly. It became the format of the download era because it shrank music files enough to share over slow connections.

AAC, Advanced Audio Coding, was standardized later in the 1990s as part of MPEG-2 and then extended in MPEG-4. It uses the same broad approach but with more flexible tools: finer control over how audio is split into frequency bands, better handling of sharp sounds, and more efficient coding of what remains. It was intended as MP3's successor, and it became the default audio codec for MP4 video, many streaming services and Apple's ecosystem.

Both are lossy. Neither can restore what was discarded, and both get worse each time a file is decoded and encoded again. If you need the background on how lossy codecs decide what to drop, the explainer on what an audio codec is covers it.

Efficiency: where the difference is audible

The practical difference between the two shows up most at lower bitrates. As the bitrate drops, an MP3 encoder has to cut harder, which typically means a lower frequency ceiling, a swishy or watery texture on cymbals and sibilants, and smeared transients. AAC tends to degrade more gracefully at the same bitrate, so a voice or a piece of music keeps more of its clarity.

At higher bitrates the gap narrows until, for most people on most equipment, it disappears. A well-encoded MP3 at a high bitrate and an AAC file at a similar setting are both commonly described as transparent for typical listening. Claims that one is always better than the other usually come from comparisons at a single bitrate, often a low one.

AAC also has profiles aimed at very low bitrates. The common profile is AAC-LC (low complexity). HE-AAC adds a technique that reconstructs high frequencies from a small amount of side information, and HE-AAC v2 adds a compact way to code stereo. These profiles suit streaming radio and voice at small sizes, though not every decoder supports them, and their reconstruction can sound artificial on some material.

Compatibility and where each plays

MP3

Plays almost everywhere: browsers, phones, desktop players, car stereos, older portable players, embedded devices, podcast apps.

AAC (LC)

Plays on current phones, browsers, smart TVs and most desktop software; it is the usual audio inside MP4 video.

HE-AAC

Widely supported on modern platforms, but older or simpler players may play it badly or not at all.

Editing apps

Both import into mainstream editors; some tools prefer to convert lossy imports to an uncompressed working format.

MP3's advantage is the long tail. If your audience includes people with old hardware, a hardware player in a classroom or a car stereo that only reads MP3 from a USB stick, MP3 removes the guesswork. For anything played on a current phone or in a browser, AAC is just as safe.

Containers and file extensions

A codec is not a file type. MP3 audio usually lives in a bare .mp3 file with ID3 tags for title, artist and artwork. AAC appears in several wrappers:

• .m4a: AAC inside an MP4-family container, typical of iPhone voice memos, iTunes purchases and many podcast exports.
• .aac: a raw AAC stream, often in a framing format called ADTS, with limited room for metadata.
• .mp4, .mov and .mkv: AAC as the audio track of a video file.
• Streaming formats such as HLS segments, where AAC is the common audio choice.

This matters when a tool asks for "AAC" but you have an M4A, or when a player shows an M4A as audio-only video. The audio inside can be identical; only the wrapper differs. Note that an M4A can also hold Apple Lossless, which is not AAC at all.

Licensing history in brief

MP3 was covered by patents for many years, which is one reason open alternatives such as Vorbis and later Opus were developed. The key MP3 patents have since expired and the main licensing program ended around 2017, so MP3 encoders now ship freely in most software.

AAC has also been licensed through a patent pool for device and software makers. For someone exporting a podcast or a lecture, this has no practical effect: you are using an encoder someone else licensed. If you are building a product that encodes or decodes AAC, check the current licensing position with a lawyer rather than relying on a blog summary.

What matters for transcription

For speech recognition, the codec matters much less than the recording. Most recognizers convert audio to mono at a low sample rate such as 16 kHz before analyzing it, which discards much of what separates a good MP3 from a good AAC file. A clear voice close to the microphone transcribes well in either codec; an echoing room transcribes poorly in both.

Codec choice starts to matter at the bottom of the bitrate range, where consonants such as s, f and th blur together. Here AAC's efficiency gives it an edge at the same size, but the real fix is not to go that low. The article on audio bitrate for speech recognition discusses where that floor sits, and the comparison of MP3 and WAV for transcription covers lossless versus lossy, which this page leaves aside.

Choosing by situation

• Publishing a podcast or audio download for the widest audience: MP3 at a moderate bitrate is the low-risk default, and AAC in M4A is equally fine if your host and audience use current apps.
• Audio for video: AAC, since it is the expected track format inside MP4.
• Very small files for voice, such as long recordings on limited storage: AAC (or Opus where supported) holds up better than MP3 at the same size.
• Sending a file to someone with old hardware or an unknown player: MP3.
• Transcribing: whichever file you already have. Do not convert between them first.
• Archiving or editing: neither. Keep a lossless master and export lossy copies from it.

Mistakes and trade-offs to watch

• Converting AAC to MP3 or MP3 to AAC "for compatibility" when the destination accepts the original. Every conversion between lossy codecs adds new artifacts on top of the old ones.
• Re-encoding at a higher bitrate expecting better sound. A 64 kb/s file converted to 256 kb/s is bigger, not clearer.
• Comparing at one bitrate and generalizing. A test at very low bitrate says little about high-bitrate exports.
• Choosing HE-AAC for music that will play on unknown devices. Its reconstruction can misbehave on players that only partly support it.
• Forgetting that the extension may lie. If a file will not play, inspect what codec is actually inside before converting.
• Assuming gapless albums or seamless loops will survive. Both codecs add a little padding at the start and end of a file, and gapless playback depends on metadata the player must honor.

How mydubly handles AAC and MP3

mydubly's audio to text tool accepts MP3, AAC and M4A files, along with WAV, OGG, FLAC and video files, up to two hours each. Whatever codec you upload, the browser decodes it on your device, converts it to 16 kHz mono, splits it into chunks of about 30 seconds and compresses each chunk to Opus at about 32 kb/s before sending it over HTTPS for recognition with Whisper. So an AAC file has no transcription advantage over an MP3 of similar quality, and neither costs a large upload.

Pricing does not depend on the codec: 1 credit per minute with a 5-credit minimum, so a 50-minute lecture costs 50 credits (5¢). You get a plain transcript, a timestamped transcript, and SRT and VTT subtitles. The M4A to text page covers the phone-recording case specifically.

When mydubly dubs a video or audio file, the audio it produces is mono AAC: the translated voice mixed over the source's music and effects, with the original speech removed.

A sensible default

If you only remember one thing: export AAC when the audio goes into video or onto current devices, export MP3 when you cannot predict the player, and never convert one into the other without a reason. For transcription, upload the file you have, as close to the original recording as possible, and check a few minutes of the result against the audio.