The Complete Guide to Audio File Formats

The audio format maze

You record a voice memo on your phone and it comes out as an M4A. You download a song and it is an MP3. A friend sends you a FLAC and your music player has never heard of it. Someone in a forum tells you to "just use OGG" and you have no idea why. The world of audio file formats is genuinely confusing, and most people work around it by sticking with whatever their device hands them.

That works, until it doesn't. The wrong format can quietly ruin a podcast, balloon a folder of recordings to ten times the size it needs to be, or hand you a file your editor cannot open. Picking well is not hard once you understand what each format is actually doing under the hood, and that is what this guide is about.

We will start with the formats themselves in plain language, move into the practical decisions most people actually face, and finish with a more technical look at how lossy compression works, why bitrate matters, and what the research on perceptual coding actually says. By the end you should be able to pick the right format for any job and explain why to anyone who asks.

The formats, in plain English

WAV (Waveform Audio File Format)

WAV is the simplest answer to the question "how do I save audio on a computer". It stores the sound exactly as the microphone heard it, sample by sample, with no compression at all. Open a WAV file from 1995 today and it plays back identically to the day it was recorded.

The trade-off is size. A minute of CD-quality stereo WAV is roughly 10 MB. An hour of it is around 600 MB. That is fine for editing in a digital audio workstation, where you want pristine source material, but it is the wrong format for sharing or streaming. Email providers will reject it. Cloud storage will fill up. Phones will refuse to play it back smoothly.

Use WAV when you are recording, editing, or archiving. Convert to something else before you publish.

MP3 (MPEG-1 Audio Layer III)

MP3 is the format that taught the world that audio could be small. It came out of a research effort at the Fraunhofer Institute in the early 1990s, and it works by throwing away the parts of the sound that human ears cannot reliably detect. At 128 kbps, a typical pop song takes around 1 MB per minute, which is roughly a tenth the size of the WAV original.

The catch, and there is one, is that the information is gone. Re-encoding an MP3 into another MP3 loses a little more each time, the same way photocopying a photocopy degrades the image. For listening, this rarely matters. For editing, archiving, or running through further processing, it does.

MP3 is the right answer when you need universal compatibility. Every phone, every car stereo, every web browser, every podcast app understands it. No other format comes close on that metric.

AAC and M4A (Advanced Audio Coding)

AAC is what MP3 grew up to become. It is the default audio codec for the iTunes Store, Apple Music, Spotify, YouTube, and broadcast television in most of the world. At the same file size as MP3 it sounds noticeably better, and at the same perceived quality it is roughly 20 to 30 percent smaller.

M4A is just AAC wrapped in Apple's MP4 container, which is why iPhones produce M4A by default. The audio inside is identical to AAC in any other container.

Modern devices play AAC without issue. Very old hardware, some car stereos from the early 2000s, and a few legacy players still want MP3, which is the main reason MP3 has not been retired.

FLAC (Free Lossless Audio Codec)

FLAC is the answer to "I want compression without losing anything". It works the same way a ZIP file works: it analyses the data, finds patterns, and stores them more efficiently, but every bit can be perfectly reconstructed. Decoding a FLAC and re-encoding it back to WAV gives you a file that is byte-for-byte identical to the original.

The cost is that FLAC files are still relatively large, around 6 to 8 MB per minute, depending on the music. That is roughly half the size of WAV but still much larger than MP3 or AAC. Support is now broad on desktop and most modern phones, but older car systems and some streaming devices still ignore it.

FLAC is the right format for archiving music and recordings you want to keep forever. It is overkill for casual listening.

OGG Vorbis and Opus

OGG is a container format, and what most people call "OGG" is actually Vorbis or Opus audio inside an OGG container. Both are open, royalty-free codecs that often outperform MP3 and AAC at similar bitrates.

Opus in particular is one of the strongest general-purpose audio codecs available today. It scales from very low bitrates suitable for voice calls all the way up to high-fidelity music, and it powers the audio side of WhatsApp voice notes, Zoom, Discord, and most modern WebRTC applications.

The catch is consumer support. Plenty of devices and editors still do not handle OGG or Opus natively. If you control the playback environment, a web app, a game, an internal tool, these codecs are an excellent choice. If you are handing the file to a non-technical friend, they are not.

AIFF, WMA, and the rest

A handful of other formats appear from time to time. AIFF is Apple's equivalent of WAV, uncompressed and used inside the macOS audio ecosystem. WMA is Microsoft's lossy format, mostly historical now. ALAC is Apple's lossless format, comparable to FLAC, and is sometimes used inside Apple Music for high-resolution streams. For almost any practical purpose, the five formats above cover what you need.

Quick decision guide

A short reference for the most common situations:

• Editing audio in a DAW? Work in WAV. Export to your delivery format.
• Archiving music or important recordings? FLAC.
• Sharing a voice memo with a friend? M4A or MP3 at 128 kbps.
• Producing a podcast? MP3 at 128 to 192 kbps for the final delivery.
• Embedding audio on a website? AAC (in MP4) with an OGG/Opus fallback.
• Voice calls or real-time audio? Opus, almost always.
• Maximum compatibility, no questions asked? MP3.

If you only remember one rule: archive in lossless, distribute in lossy, and never edit a lossy file you intend to publish again.

Convert without uploading

TabTasker's Audio Converter lets you convert audio to MP3 or WAV locally in your browser using the built-in Web Audio API. No uploads, no limits, no accounts. Drop your file in, pick the output format, and download. The conversion happens on your device.

If you also need to shrink files for sharing, the Audio Compressor handles that. To clean up background hiss, try the Audio Denoiser. For cutting clips down to size, the Audio Trimmer does it all without leaving the tab.

Bitrate: the single most important number for lossy formats

For any lossy format (MP3, AAC, OGG Vorbis, Opus, WMA), bitrate is the lever that controls the trade-off between file size and audio quality. Bitrate measures how many kilobits of compressed data are stored per second of audio, and it is almost always written as kbps.

A rough map of where the useful thresholds sit:

• 32 to 64 kbps: Telephony quality. Intelligible voice, no music fidelity. Useful for archived voice notes and bandwidth-constrained streaming.
• 96 kbps: Spoken content. Fine for podcasts, audiobooks, and interviews. Music will sound thin.
• 128 kbps: The historical "good enough" setting. Acceptable on everyday speakers and earbuds, audibly compressed on better equipment.
• 192 to 256 kbps: High quality. For most listeners on most equipment, indistinguishable from the original in casual listening.
• 320 kbps: Maximum MP3 quality. Effectively transparent to almost everyone outside a controlled listening test.

Two refinements are worth knowing. First, AAC achieves comparable perceived quality at roughly 20 to 30 percent lower bitrate than MP3, and Opus does even better, especially below 96 kbps. A 96 kbps Opus stream often sounds better than a 128 kbps MP3. Second, modern encoders default to variable bitrate (VBR), which spends more bits on complex passages and fewer on simple ones, producing better quality at the same average size. Unless you have a specific reason, prefer VBR.

Lossless vs. lossy: when the difference actually matters

The debate around lossless audio is older than streaming, but the engineering answer has been stable for years.

In double-blind listening tests, the overwhelming majority of listeners cannot reliably distinguish a well-encoded 256 kbps AAC file from its lossless original. This finding has been replicated many times, with trained engineers and casual listeners alike, on studio monitors and consumer headphones. Human hearing has hard physical and perceptual limits, and modern lossy encoders sit very close to those limits.

Lossless formats still matter, but for narrower reasons than audiophile marketing suggests:

1. Editing and processing. Every time a lossy file is decoded, modified, and re-encoded, it loses a small amount of quality. Editing in lossless avoids that generational loss entirely.
2. Archival. A lossless master can be transcoded into any future format at any quality. A lossy master is permanently constrained to its encoding choices.
3. Critical mastering work. When you are mastering audio for distribution, you want every dB of headroom and every bit of resolution. Lossless is the only sensible choice.

For final distribution and everyday listening, a high-bitrate AAC, Opus, or MP3 file is, in almost every case, audibly sufficient.

Choosing the right format for your workflow

The best format depends much more on where the audio is going than on how it sounds in isolation. A few common workflows:

• Web playback. AAC inside an MP4 container, with OGG/Opus as a fallback, gives the broadest browser coverage at the best quality per byte.
• Mobile apps. AAC on iOS, AAC or Opus on Android. Both platforms include high-quality decoders by default.
• Podcast distribution. MP3 at 128 to 192 kbps remains the standard. Some platforms now accept AAC, but MP3 is still the safe bet.
• Music streaming. Services have settled on AAC, with FLAC and ALAC for "lossless" tiers.
• Game audio. Opus for music and ambient loops, often with uncompressed PCM (WAV) for short sound effects where decode latency matters.
• Broadcast and post-production. WAV or BWF (Broadcast Wave Format) is the universal interchange standard.

Convert your files locally to whichever format your destination needs, without sending them to a server. The conversion runs entirely in your browser.

How lossy compression actually works: a closer look

The everyday answer above is enough for most decisions. For readers who want to understand why MP3 and AAC can throw away most of a signal without sounding obviously broken, the next sections look at the underlying engineering.

Perceptual coding and psychoacoustics

Lossy audio codecs are built on a branch of acoustics called psychoacoustics, the study of how the ear and brain process sound. The central observation is that human hearing is not a flat measurement instrument. We are very sensitive to mid-range frequencies (roughly 2 to 5 kHz, the range of speech), less sensitive to high and low extremes, and we exhibit several forms of perceptual masking where loud sounds at one frequency render quieter sounds at nearby frequencies inaudible.

A perceptual coder exploits these gaps. It models, mathematically, what the listener is and is not capable of hearing in each short slice of the audio, then spends bits only on the parts that fall above the masking threshold. Everything below the threshold is discarded. To the listener, the result sounds essentially identical, because the discarded information was, by definition, inaudible in context.

This is why bitrate matters so much. A higher bitrate gives the coder more bits to spend on the parts of the signal that are near the masking threshold, where small errors might just barely be perceptible. A lower bitrate forces the coder to be more aggressive, eventually crossing the threshold and producing audible artefacts: the watery, swishy sound of "MP3 distortion" you may have heard at very low bitrates.

The transform: from time to frequency

Almost every modern lossy codec starts by transforming a short window of audio from the time domain (a stream of samples) into the frequency domain (a set of coefficients for different frequencies). MP3 uses a hybrid filterbank built around a Modified Discrete Cosine Transform (MDCT). AAC uses a pure MDCT. Opus uses both an MDCT path (for music) and a linear predictive path (for speech), and switches between them dynamically.

The frequency-domain representation is where psychoacoustic masking is easiest to apply, because masking is itself a frequency-domain phenomenon. Once the signal is decomposed into frequency components, the coder can identify which components are perceptually important and quantise them with high precision, and which are masked and can be quantised coarsely or dropped entirely.

Quantisation and entropy coding

After the transform, the codec quantises each frequency coefficient, mapping a continuous value onto a discrete integer. Quantisation is where the lossy part of "lossy compression" actually happens. Coefficients that fall below the masking threshold are quantised to zero. Coefficients above the threshold are quantised to whatever precision the bit budget allows.

The quantised integers are then run through an entropy coder, typically Huffman coding in MP3 and AAC, or range coding in Opus. Entropy coding is lossless and shrinks the integer stream by exploiting statistical redundancy: common values get short codes, rare values get long codes. This is the same family of techniques that powers ZIP and PNG.

The result is a compact bitstream that, when decoded, runs the inverse transform and produces a waveform very close to the original. The only differences are those introduced by quantisation, and the coder's job is to ensure those differences fall below the listener's perceptual threshold.

Why AAC and Opus outperform MP3

AAC improves on MP3 in several specific ways. It uses longer transform windows, which give better frequency resolution; it supports more sophisticated joint stereo coding, which exploits redundancy between left and right channels; and its psychoacoustic model is more refined, particularly for transient sounds like percussion. The net effect is roughly a 20 to 30 percent reduction in bitrate for the same perceived quality.

Opus goes further by combining ideas from two earlier codecs, SILK (originally developed by Skype for voice) and CELT (a low-latency music codec). At low bitrates it leans on SILK's linear-predictive approach, which models the resonances of the human vocal tract very efficiently. At higher bitrates it shifts to CELT's MDCT-based approach, which handles music well. The switching is seamless, and the result is a single codec that performs at or near the state of the art across the entire range from 6 kbps voice to 256 kbps music.

Variable bitrate, constant bitrate, and average bitrate

Bitrate can be controlled in three broad ways. Constant bitrate (CBR) spends the same number of bits per second regardless of content, which is useful for streaming over fixed-bandwidth channels but wastes bits on simple passages and starves complex ones. Variable bitrate (VBR) lets the encoder decide how many bits each section needs, producing the best quality per average byte. Average bitrate (ABR) is a hybrid that targets a specific average while allowing local variation.

For almost any modern use case, VBR is the right choice. Streaming services that historically demanded CBR have moved to adaptive bitrate streaming, which switches between several precomputed VBR encodings depending on the listener's connection.

Bitrate-distortion theory in one paragraph

Information theory provides a hard floor on how much a signal can be compressed without loss. For audio, that floor is set by the signal's entropy: the statistical complexity of the waveform. Lossless coders like FLAC approach this floor but cannot go below it. Lossy coders break through it by accepting controlled distortion, and the field of rate-distortion theory studies the optimal trade-off curve between bitrate and a chosen distortion metric. Psychoacoustic coders are essentially rate-distortion optimisers with a distortion metric tuned to human hearing. The reason they work so well is that human hearing turns out to be much less sensitive than the raw signal entropy suggests, leaving a large region of "free" compression that perceptual coding can exploit.

A short forecast

A few trends are worth watching over the next several years.

First, Opus continues to gain ground. It is already dominant in real-time communication and is increasingly common in web audio. Music streaming services have been slower to adopt it, partly for historical compatibility reasons, but the technical case for Opus over AAC is strong and pressure is building.

Second, the rise of on-device machine learning is producing a new generation of neural audio codecs (Lyra, EnCodec, SoundStream) that achieve dramatically lower bitrates for speech and even music, by replacing handwritten signal-processing pipelines with learned models. These codecs are not yet in everyday consumer use, but they are likely to appear first in messaging apps, then in broadcast, and eventually in general distribution.

Third, lossless distribution is becoming cheaper and more common. Bandwidth costs have fallen far enough that streaming a FLAC or ALAC file is no longer expensive, and "lossless" has become a marketing-grade differentiator. For most listeners on most equipment it remains audibly indistinguishable from high-bitrate lossy, but the perception of quality matters, and the engineering cost of providing lossless tiers is now low.

The bottom line

Audio formats are not as complicated as the variety of acronyms suggests. There are uncompressed formats (WAV, AIFF) for editing and archival, lossless compressed formats (FLAC, ALAC) for keeping originals in less space, and lossy formats (MP3, AAC, Opus) for sharing and streaming. Bitrate, not format, is what most directly controls the quality of a lossy file. And for almost all everyday listening, modern lossy codecs at moderate bitrates are perceptually indistinguishable from the source.

Pick the format that fits the destination, not the one that sounds most impressive on paper. Archive in lossless, distribute in lossy, and let the codec do what decades of research have made it very good at: removing the information your ear was never going to use anyway.