How MP3 Files Work: Frames, Bitrate and ID3 Tags

The Layout of a Typical MP3 File

You can see this for yourself in a binary file viewer: a tagged MP3 begins with the ASCII letters ID3, one of the signatures listed in file signatures and magic numbers.

Frames: The Building Blocks

The Frame Header

Each frame starts with a four-byte header. Its first 11 bits are all set to 1, a sync word that lets a decoder find frames even if it starts reading mid-file. The rest of the header specifies:

• MPEG version and layer (MPEG-1 Layer III is the classic "MP3").
• Bitrate for this frame, such as 128 or 320 kbps.
• Sample rate: 44.1 kHz, 48 kHz or 32 kHz for MPEG-1; lower rates for MPEG-2 and 2.5.
• Padding, a flag that adds a byte to keep the average bitrate exact.
• Channel mode: stereo, joint stereo, dual channel or mono.

Samples per Frame

An MPEG-1 Layer III frame always represents 1,152 samples per channel. At 44.1 kHz, that's about 26.1 milliseconds of sound. A three-minute song therefore contains roughly 6,900 frames.

Frame Size

The frame's size in bytes follows from its bitrate and sample rate:

frame bytes = 144 × bitrate ÷ sample rate (+ 1 if padded)
128 kbps at 44.1 kHz: 144 × 128000 ÷ 44100 ≈ 417 or 418 bytes

Because frames are self-describing, tools can walk through a file frame by frame without decoding the audio. That's exactly how lossless splitters work, as explained in how to split an MP3 without losing quality.

Bitrate: CBR, VBR and ABR

Bitrate is how many bits per second the audio uses. Higher bitrates keep more detail.

• CBR (constant bitrate): every frame uses the same bitrate. Sizes and durations are easy to calculate.
• VBR (variable bitrate): each frame uses as many bits as the encoder thinks it needs. Quiet or simple passages use fewer bits; complex passages use more. You get better quality for the size, but duration can't be calculated from file size alone.
• ABR (average bitrate): varies per frame but targets an average.

The Xing / Info Header

In a VBR file, a player can't estimate the duration from the file size. So encoders write a special first frame, containing no audible audio, with a Xing tag (for VBR) or Info tag (for CBR). It stores the total number of frames and bytes, and a table of contents for seeking. The LAME encoder adds extra fields, including encoder delay and padding, which gapless players use to remove the silence encoders add at the start and end.

When a file is split, each part needs its own correct header. Without one, a player may show the original file's length, or seek inaccurately.

ID3 Tags

ID3 tags hold metadata that the MP3 audio format itself has no place for.

• ID3v1 is a fixed 128-byte block at the end of the file with short fields for title, artist, album, year, comment and genre.
• ID3v2 sits at the start and is far more flexible: long text, multiple languages, embedded images, lyrics, and even chapter markers. Versions 2.3 and 2.4 are the common ones.

Fields you'll care about when splitting or organising audio include title (TIT2), artist (TPE1), album (TALB) and track number (TRCK). A splitter that updates title and track number for each part saves you renaming everything in a tag editor.

How MP3 Compression Discards Sound

The encoder splits audio into frequency bands and uses a psychoacoustic model to decide which components are masked by louder nearby sounds or are otherwise unlikely to be heard. Those get fewer bits or none. This is why MP3 is lossy: decoding never returns the exact original samples, and encoding again discards more. The MP3 splitter avoids that second loss by never decoding the audio: it copies frames and writes fresh headers and tags for each part.

Summary

An MP3 is tags plus a stream of self-describing frames, each about 26 ms long at 44.1 kHz. Bitrate can be constant or variable, a Xing/Info header tells players the length, and ID3 tags carry the metadata.