Lossy or lossless: which audio format should I keep my library in?

For long-term archiving choose FLAC (lossless, ~50% the size of WAV). For everyday listening on phones and streaming use MP3 320 kbps or Opus 192 kbps (transparent quality, no audible difference at normal volumes). Avoid converting lossy → lossy → lossy: every step compounds artifacts.

Lossy or lossless: which audio format should I keep my library in?

For long-term archiving choose FLAC (lossless, ~50% the size of WAV). For everyday listening on phones and streaming use MP3 320 kbps or Opus 192 kbps (transparent quality, no audible difference at normal volumes). Avoid converting lossy → lossy → lossy: every step compounds artifacts.

Guide

MP3 Format: Complete Technical Guide to MPEG Audio Layer III

PC By Pablo Cirre • Updated May 7, 2026

What Is MP3 Format?

MP3 (MPEG-1 Audio Layer III) is the most widely used lossy audio compression format in the world. Developed primarily by the Fraunhofer Society in Germany in collaboration with AT&T Bell Labs, Philips, and others, MP3 was standardized as part of MPEG-1 (ISO/IEC 11172-3) in 1993. The format's patents expired in 2017, making it fully royalty-free.

MP3 reduces audio file size by 75–90% compared to uncompressed PCM audio (CD quality) while maintaining perceptually acceptable quality. A 5-minute song on a CD requires about 50 MB of raw PCM data; the same song as a 192 kbps MP3 is approximately 7 MB.

How MP3 Compression Works

MP3 uses a psychoacoustic model — a mathematical model of human hearing — to identify and discard audio information that the human ear is unlikely to perceive. This is fundamentally different from lossless compression (which reduces file size without discarding any data).

The MP3 Encoding Pipeline

PCM Input Audio
    │
    ▼
┌─────────────────────────────────────┐
│ Polyphase Filterbank (PQF)          │
│ Splits audio into 32 frequency      │
│ sub-bands                           │
└──────────────┬──────────────────────┘
               │
               ▼
┌─────────────────────────────────────┐
│ MDCT (Modified Discrete Cosine      │
│ Transform) — 576-point transform    │
│ yields 576 frequency coefficients   │
└──────────────┬──────────────────────┘
               │
               ├─────────────────────────────┐
               ▼                             ▼
┌──────────────────────────┐    ┌────────────────────────────┐
│ Psychoacoustic Model     │    │ Quantization & Huffman      │
│ - Masking threshold      │───▶│ Coding                      │
│ - Absolute threshold     │    │ - Coeff quantized per model │
│ - Spectral masking       │    │ - Huffman entropy coding    │
│ - Temporal masking       │    └────────────┬───────────────┘
└──────────────────────────┘                 │
                                             ▼
                                   ┌─────────────────┐
                                   │ MP3 Bitstream   │
                                   │ (frames)        │
                                   └─────────────────┘

Psychoacoustic Masking

The psychoacoustic model exploits two properties of human hearing:

Frequency masking: A loud sound at one frequency makes quieter sounds at nearby frequencies imperceptible. If a 1 kHz tone is played loudly, a quieter tone at 1.1 kHz is masked (hidden) and can be discarded.
Temporal masking: After a loud sound, the ear needs time to recover sensitivity. Quiet sounds occurring just before (~20ms) or just after (~200ms) a loud transient are masked and can be discarded.

By identifying which audio information falls below the masking threshold at each moment, MP3 can allocate more bits to perceptually important sounds and fewer (or zero) bits to masked sounds.

MDCT Transform

The 576-point Modified Discrete Cosine Transform converts a block of time-domain audio samples into 576 frequency-domain coefficients. These coefficients represent the amplitude of each frequency component in that time window. This frequency-domain representation is necessary for the psychoacoustic model to compare energy levels across frequency bands.

For transient sounds (drum hits, plosive consonants), MP3 switches to a short block mode (3 × 192 samples) to reduce pre-echo artifacts — a ringing sound that appears before a transient in incorrectly encoded MP3s.

MP3 Frame Structure

MP3 audio is organized into frames. Each frame is an independent compressed block covering 1152 PCM samples:

MP3 Frame (1152 samples / sample_rate duration)
├── Header (4 bytes)
│   ├── Sync word: 0xFFF (11 bits) — marks frame start
│   ├── MPEG version (1=MPEG1, 0=MPEG2)
│   ├── Layer (11 = Layer 3)
│   ├── Bitrate index (4 bits → lookup table)
│   ├── Sample rate index (2 bits: 44100/48000/32000 Hz)
│   ├── Padding flag
│   ├── Channel mode (stereo/joint stereo/dual/mono)
│   └── Mode extension (for joint stereo)
├── Side information (17/32 bytes)
│   ├── Main data begin pointer
│   ├── Granule 1 information (scalefactor bands, Huffman table selection)
│   └── Granule 2 information
└── Main data (variable length)
    ├── Scalefactors
    ├── Huffman-coded frequency coefficients
    └── Ancillary data (optional)

Frame size in bytes = 144 × bitrate / sample_rate (+ padding byte if padding flag set)

For 128 kbps at 44100 Hz: 144 × 128000 / 44100 = 417 bytes per frame (26 ms of audio)

Bitrate: CBR vs. VBR vs. ABR

Constant Bitrate (CBR)

Every frame uses exactly the same bitrate. Simple, universally compatible, predictable file size. Wastes bits on easy passages (silence) while potentially having insufficient bits for complex passages.

Variable Bitrate (VBR)

The encoder allocates more bits to complex audio passages and fewer to simple ones, targeting a specified quality level. VBR produces better quality-per-byte ratios but:

File size is unpredictable
Some old hardware players had VBR seeking issues
Specified as a quality level (e.g., LAME -V 0 to -V 9, where 0 = best quality)

LAME VBR quality levels (approximate):

LAME -V	Avg bitrate	Typical use
V0	~245 kbps	Transparent quality — preferred for archival
V2	~190 kbps	Excellent — virtually transparent
V5	~130 kbps	Good — acceptable for casual listening
V9	~65 kbps	Low quality — streaming only

Average Bitrate (ABR)

A compromise: targets a specific average bitrate while allowing frame-by-frame variation. More predictable file size than VBR while better quality distribution than CBR.

Recommended Bitrates

Use case	Recommended bitrate
Music archival	LAME VBR V0 (~245 kbps) or 320 kbps CBR
High-quality music	256 kbps CBR or VBR V2
General music	192 kbps CBR
Podcast / speech	64–128 kbps CBR (mono for speech)
Low-bandwidth streaming	32–64 kbps

Stereo Encoding Modes

MP3 supports multiple stereo encoding strategies:

Stereo (True Stereo): left and right channels encoded independently. Best for music with true stereo information.
Joint Stereo / Mid-Side (MS) Stereo: encodes the sum (mid = L+R) and difference (side = L-R) signals instead of L and R directly. More efficient because most music has correlated channels — the mid signal carries most energy and gets more bits.
Intensity Stereo: at low bitrates, encodes only a single mono spectrum for high frequencies but preserves directional cues via scale factors. Artifacts are audible at high bitrates.
Dual Channel: two independent mono channels (for bilingual audio).

ID3 Tags: Metadata in MP3

MP3 files can carry metadata via ID3 tags:

ID3v1 (legacy)

Fixed 128-byte trailer at the end of the file
Fields: Song title (30 chars), Artist (30), Album (30), Year (4), Comment (30), Genre (1 byte index)
No Unicode support; fields truncated to fixed lengths

ID3v2 (modern, near-universal)

Variable-length header at the beginning of the file
Flexible frame-based structure (each frame has a 4-character code + size + data)
Supports Unicode (UTF-8, UTF-16)
Common frames:
- TIT2 — Title
- TPE1 — Lead artist
- TALB — Album
- TDRC — Recording date
- TRCK — Track number
- APIC — Attached picture (album art, any image format)
- COMM — Comments
- TCON — Genre
- TLEN — Length in milliseconds
- TPOS — Part of set (disc number)
- TXXX — User-defined text frame

MP3 Quality Artifacts

At low bitrates, MP3 produces characteristic artifacts:

Pre-echo: Ringing before a transient (drum hit). Caused by MDCT spreading energy backward in time.
Ringing/smearing: Musical notes have audible "echoes" or tonal artifacts.
Low-frequency pumping: At very low bitrates, bass energy varies unnaturally.
Metallic/watery sound: Common at 64–128 kbps; caused by extreme quantization of high-frequency content.
Mosquito noise: High-frequency noise around transients, particularly noticeable on cymbals and voices at low bitrates.

Modern encoders (LAME 3.100+) minimize these artifacts significantly compared to early 1990s encoders.

LAME: The Gold Standard MP3 Encoder

LAME (LAME Ain't an MP3 Encoder) is the most widely used and highest-quality open-source MP3 encoder. Key encoding commands:

# VBR V0 (highest quality VBR)
lame -V 0 input.wav output.mp3

# 320 kbps CBR (maximum quality)
lame -b 320 input.wav output.mp3

# 192 kbps CBR with ID3 tags
lame -b 192 --tt "Song Title" --ta "Artist" --tl "Album" input.wav output.mp3

# Podcast quality (mono, 64 kbps)
lame -b 64 -m m input.wav output.mp3

# VBR V2 with joint stereo (recommended default)
lame -V 2 --preset fast standard input.wav output.mp3

MP3 vs. Alternative Audio Formats

Format	Type	Quality at 128 kbps	Compatibility	Use case
MP3	Lossy	Good	Universal	General distribution
AAC	Lossy	Better	Very wide	Apple, streaming
OGG Vorbis	Lossy	Better	Web/desktop	Games, web audio
Opus	Lossy	Excellent	Browsers, VoIP	Streaming, calls
FLAC	Lossless	Perfect	Desktop/HiFi	Archival, audiophile
WAV/AIFF	Uncompressed	Perfect	Universal	Production, editing

MP3's main advantage is its absolute universal compatibility — it plays on every device, platform, and software ever made. For quality-sensitive applications, AAC or Opus are better choices. For archival, FLAC is preferred.

Related conversions

Audio format pairs that come up most often:

Related conversions

Put what you just learned into practice — convert your files now in seconds, free and without registration.

Convert →

Convert →

Convert →

Convert →

Convert →

Convert →

Frequently Asked Questions

For music you want to keep indefinitely, LAME VBR V0 (averaging ~245 kbps) or 320 kbps CBR are the best choices — both are considered "transparent" quality, meaning most listeners cannot distinguish them from the original CD in a blind test. For everyday listening where storage matters, 192–256 kbps CBR strikes an excellent balance. For podcasts or speech content, 64–128 kbps is sufficient since speech is less complex than music. Avoid going below 128 kbps for music if quality matters — artifacts like metallic "watery" sound become audible.

For music you querer keep indefinitely, LAME VBR V0 (averaging ~245 kbps) ou 320 kbps CBR are the best choices — both are considered "transparent" quality, meaning most listeners cannot distinguish them de the original CD em um blind test. para everyday listening where storage matters, 192–256 kbps CBR strikes an excellent balance. para podcasts ou speech content, 64–128 kbps is sufficient since speech is less complexo than music. Avoid going below 128 kbps para music if quality matters — artifacts like metallic "watery" sound become audible.

For music you want to keep indefinitely, LAME VBR V0 (averaging ~245 kbps) oder 320 kbps CBR are the best choices — both are considered "transparent" quality, meaning most listeners cannot distinguish them von the original CD in einem blind test. für everyday listening where storage matters, 192–256 kbps CBR strikes an excellent balance. für podcasts oder speech content, 64–128 kbps is sufficient since speech is less complex than music. Avoid going below 128 kbps für music if quality matters — artifacts like metallic "watery" sound become audible.

For music you want to keep indefinitely, LAME VBR V0 (averaging ~245 kbps) o 320 kbps CBR are the best choices — both are considered "transparent" quality, meaning most listeners cannot distinguish them de the original CD en un blind test. para everyday listening where storage matters, 192–256 kbps CBR strikes an excellent balance. para podcasts o speech content, 64–128 kbps is sufficient since speech is less complex than music. Avoid going below 128 kbps para music if quality matters — artifacts like metallic "watery" sound become audible.

For long-term archiving choose FLAC (lossless, ~50% the size of WAV). For everyday listening on phones and streaming use MP3 320 kbps or Opus 192 kbps (transparent quality, no audible difference at normal volumes). Avoid converting lossy → lossy → lossy: every step compounds artifacts.

MP3 is the most universally compatible audio format, but not the highest quality for a given file size. AAC provides noticeably better quality than MP3 at equivalent bitrates and is the default format on Apple platforms and YouTube. Opus is technically superior to both for streaming and voice applications. OGG Vorbis offers better quality than MP3 at equal bitrates. For archiving, FLAC (lossless) is always preferred over any lossy format. MP3 remains the best choice when maximum device compatibility is the primary concern — if you need audio that will play on any device, any car stereo, any old media player, MP3 is still the safest choice.

MP3 é o formato de áudio com maior compatibilidade universal, mas não o de maior qualidade para um tamanho dado. AAC oferece qualidade visivelmente melhor que MP3 em bitrates equivalentes e é o formato padrão em plataformas Apple e YouTube. Opus é tecnicamente superior a ambos para streaming e aplicações de voz. OGG Vorbis oferece melhor qualidade que MP3 em bitrates iguais. Para arquivamento, FLAC (lossless) é sempre preferido sobre qualquer formato com perdas. MP3 segue sendo a melhor escolha quando compatibilidade máxima de dispositivo é a prioridade — se você precisa de áudio que toque em qualquer dispositivo, qualquer som de carro ou reprodutor antigo, MP3 segue sendo a opção mais segura.

MP3 ist das Audioformat mit der größten universellen Kompatibilität, aber nicht das mit der besten Qualität für eine gegebene Größe. AAC bietet bei gleicher Bitrate spürbar bessere Qualität als MP3 und ist das Standardformat auf Apple-Plattformen und YouTube. Opus ist technisch beiden überlegen für Streaming und Sprach-Anwendungen. OGG Vorbis bietet bessere Qualität als MP3 bei gleicher Bitrate. Für Archivierung wird FLAC (verlustfrei) immer jedem verlustbehafteten Format vorgezogen. MP3 bleibt die beste Wahl, wenn maximale Geräte-Kompatibilität die Priorität ist — wenn du Audio brauchst, das auf jedem Gerät, jeder Auto-Stereoanlage oder jedem alten Media-Player abspielbar sein muss, ist MP3 weiterhin die sicherste Wahl.

MP3 es el formato de audio con mayor compatibilidad universal, pero no el de mayor calidad para un tamaño dado. AAC ofrece calidad notablemente mejor que MP3 a bitrates equivalentes y es el formato por defecto en plataformas Apple y YouTube. Opus es técnicamente superior a ambos para streaming y aplicaciones de voz. OGG Vorbis ofrece mejor calidad que MP3 a bitrates iguales. Para archivado, FLAC (lossless) siempre se prefiere sobre cualquier formato con pérdida. MP3 sigue siendo la mejor opción cuando la compatibilidad máxima de dispositivo es la prioridad — si necesitas audio que reproduzca en cualquier dispositivo, cualquier estéreo de coche o reproductor antiguo, MP3 sigue siendo la opción más segura.

Spoken-word podcasts: 44.1 kHz mono, 64–96 kbps Opus or 96 kbps MP3. Music podcasts: 44.1 kHz stereo, 128–192 kbps. Going above wastes bandwidth — speech has limited frequency content above 7 kHz, and listeners on data plans appreciate the smaller file. Apple Podcasts and Spotify both accept up to 48 kHz / 320 kbps.

CBR (Constant Bitrate) uses exactly the same bitrate for every frame of audio, resulting in a predictable file size. VBR (Variable Bitrate) allocates more bits to complex audio moments and fewer to simpler ones, achieving better average quality for the same file size. For example, a LAME VBR V2 file averaging 190 kbps will sound better than a 192 kbps CBR file at roughly the same size. VBR is generally recommended for music quality. CBR is better for streaming where constant data rate is required, and for compatibility with very old devices that had VBR seeking issues (though modern devices all handle VBR correctly).

CBR (Constant bitrate) uses exactly the same bitrate para every frame of audio, resulting em um predictable tamanho do arquivo. VBR (Variable bitrate) allocates more bits to complexo audio moments e fewer to simplesr ones, achieving better average quality para the same tamanho do arquivo. Por exemplo, a LAME VBR V2 arquivo averaging 190 kbps will sound melhor que a 192 kbps CBR arquivo at roughly the same size. VBR is Geralmente recommended para music quality. CBR is better para streaming where constant data rate is required, e para compatibilidade com very old devices that had VBR seeking issues (though dispositivos modernoos all handle VBR correctly).

CBR (Constant Bitrate) uses exactly the same Bitrate für every frame von audio, resulting in einem predictable Dateigröße. VBR (Variable Bitrate) allocates more bits to complex audio moments und fewer to einfachr ones, achieving better average quality für the same Dateigröße. Zum Beispiel, a LAME VBR V2 Datei averaging 190 kbps will sound besser als a 192 kbps CBR Datei at roughly the same size. VBR is Generell recommended für music quality. CBR is better für streaming where constant data rate is required, und für Kompatibilität mit very old devices that had VBR seeking issues (though moderne Geräte all handle VBR correctly).

CBR (Constant bitrate) uses exactly the same bitrate para every frame de audio, resulting en un predictable tamaño de archivo. VBR (Variable bitrate) allocates more bits to complex audio moments y fewer to simpler ones, achieving better average quality para the same tamaño de archivo. Por ejemplo, a LAME VBR V2 archivo averaging 190 kbps will sound mejor que a 192 kbps CBR archivo at roughly the same size. VBR is Generalmente recommended para music quality. CBR is better para streaming where constant data rate is required, y para compatibilidad con very old devices that had VBR seeking issues (though dispositivos modernoos all handle VBR correctly).

Lossy → lossy compounds quantization noise. Each encode discards the same kind of perceptual information again, multiplying artifacts. Always re-encode from a lossless master if you have one (WAV, FLAC, or the original recording). If only an MP3 is available, keep the bitrate at or above the source — never go up to "improve quality".

No. Converting a 128 kbps MP3 to 320 kbps does not restore quality — it only makes a larger file with the same quality as the 128 kbps original. Transcoding between lossy formats is always a generation loss: each conversion introduces new compression artifacts on top of existing ones. The information discarded by the first compression is permanently gone. For best results, always keep the original uncompressed source (WAV or FLAC) and encode to MP3 from that. If you only have an MP3, keep it at its original bitrate rather than re-encoding to a different bitrate.

No. convertendo a 128 kbps MP3 to 320 kbps does not restore quality — it only makes a larger arquivo com the same quality como o 128 kbps original. Transcoding between com perdas formatoos is always a generation loss: each conversion introduces new compressão artifacts on top of existing ones. The information discarded pelo first compressão is permanently gone. para best results, always keep the original não comprimido source (WAV ou FLAC) e encode to MP3 de that. If you only have an MP3, keep it at its original bitrate em vez de re-encoding para um different bitrate.

No. umwandelnd a 128 kbps MP3 to 320 kbps does not restore quality — it only makes a larger Datei mit the same quality als das 128 kbps original. Transcoding between verlustbehaftet Formate is always a generation loss: each conversion introduces new Komprimierung artifacts on top von existing ones. The information discarded by the first Komprimierung is permanently gone. für best results, always keep the original uncompressed source (WAV oder FLAC) und encode to MP3 von that. If you only have an MP3, keep it at its original Bitrate rather than re-encoding zu einem different Bitrate.

No. convirtiendo a 128 kbps MP3 to 320 kbps does not restore quality — it only makes a larger archivo con the same quality como el 128 kbps original. Transcoding between con pérdidas formatoos is always a generation loss: each conversion introduces new compresión artifacts on top de existing ones. The information discarded by the first compresión is permanently gone. para best results, always keep the original uncompressed source (WAV o FLAC) y encode to MP3 de that. If you only have an MP3, keep it at its original bitrate rather than re-encoding a un different bitrate.

Most modern tools (FFmpeg with <code>-map_metadata 0</code>, foobar2000, dBpoweramp) preserve ID3 tags and embedded cover art. Some quick-and-dirty converters strip them silently. KaijuConverter preserves tags in its <a href="/convert/flac-to-mp3">audio conversions</a>; if metadata is critical to you, verify the output with <code>ffprobe</code> before deleting the original.