"This FLAC sounds better than 320 MP3." "You need 96 kHz for studio quality." Sentences said constantly that mix physical truths with marketing. This article breaks down how digital audio actually works: what a sample rate captures exactly, what bit depth adds, where codecs really lose information, and when each thing matters.
From continuous wave to discrete numbers
Sound in the air is a continuous pressure variation. A computer cannot store continuity: it measures the wave many times per second and stores each measurement as a number. Two parameters define that conversion:
- Sample rate: how many measurements per second. 44,100 Hz = 44,100 values every second.
- Bit depth: with how much precision each value is measured. 16 bits = 65,536 possible levels.
The result is a giant list of numbers which, played in sequence through a speaker, reconstructs the original wave.
The sampling theorem: where 44,100 comes from
Nyquist-Shannon states that capturing a frequency requires sampling at least twice as fast. Human hearing reaches ~20 kHz, so the theoretical minimum is 40 kHz. The historical 44,100 Hz adds margin for transition filters (the guard band between 20,000 and 22,050 Hz).
Direct consequence, often ignored: a 44.1 kHz system cannot contain frequencies above 22.05 kHz — nor does it need to, since nobody hears them. If you try to record higher frequencies without filtering them first, they fold downward as fake low tones: aliasing, the artifact justifying the anti-aliasing filters applied before sampling.
What about 48, 96 or 192 kHz? They offer operational advantages (more headroom for pitch-shift and time-stretch in production, gentler filters) but zero extra audibility in final playback. Multiple controlled experiments have shown trained listeners cannot distinguish 44.1/16 from 192/24 blind when the pipeline is clean. Recording high and delivering at 44.1/48 makes sense; distributing at 96 kHz is wasted weight.
Bit depth: noise floor
Each bit adds ~6 dB of theoretical signal-to-noise ratio:
- 8 bits: 48 dB — audible hiss on quiet passages.
- 16 bits: 96 dB — the dynamic range from whisper to pain threshold fits entirely. It's the CD standard and enough for final consumption.
- 24 bits: 144 dB — its real utility is in recording: such a low noise floor lets you work with conservative input levels without fear of clipping or hiss. That margin is equally lost in distribution.
The professional rule summarized: record at 24 bits, deliver at 16 (or a lossy codec).
Lossy codecs: what exactly gets deleted
MP3, AAC, Ogg and Opus don't cut "the bad-sounding parts": they apply psychoacoustics. The encoder analyzes which spectral components your ear cannot perceive at each instant and removes or roughly approximates them:
- Temporal masking: a loud hit silences what plays right before and after it.
- Frequency masking: a powerful tone hides weaker neighboring tones close to its frequency.
- Bandwidth: beyond a certain point (bitrate-dependent), high frequencies are simply discarded.
At sufficient bitrate this deletion is inaudible by design. MP3 320 vs FLAC differences appear in specific cases: applause, cymbals, applause over cymbals, very long reverb tails and already-compressed material being re-compressed. With typical material and good encoders, massive blind tests show most people cannot distinguish 256 kbps from the original.
Practical bitrate guide:
- 128 kbps: acceptable for speech; music starts showing wear.
- 192 kbps: transparent for most people with AAC/Opus.
- 256-320 kbps: safe transparency zone even with MP3.
- Opus: today's most efficient codec; 256 kbps MP3 quality at less than half the bitrate. Ideal for streaming and voice.
Lossless: FLAC and friends
FLAC compresses the original PCM by 40-60% discarding nothing: decompression returns bit-for-bit what went in. It's the right format for archiving masters, editing without generational degradation, and audiophiles with disk space. ALAC is its Apple-ecosystem equivalent; WAV and AIFF are raw PCM without compression (large, useful for master exchange).
An important nuance: converting MP3 → FLAC recovers nothing. The MP3 damage is already baked into the samples; FLAC just preserves it faithfully.
Sample rates and codecs: quick decision table
| Use | Recommendation |
|---|---|
| Music distribution | FLAC 44.1/16 (or MP3 320/AAC 256 if size matters) |
| Streaming / web | Opus or AAC 128-192 |
| Podcast / voice | Opus 64-96 or AAC 96-128 |
| Recording / production | WAV 48/24 minimum |
| Master archive | FLAC 48/24 or WAV |
Conversion without surprises
Three spots ruin audio during conversion: resampling without proper filtering (aliasing), reducing bits without dithering (correlated truncation noise), and chaining lossy generations (each MP3 pass adds artifacts). Our audio converter processes everything locally in your browser with ffmpeg/WebAssembly, never uploading your files to any server — and if you work with loudness, the LUFS normalization guide explains how to level correctly before converting.
FAQ
Are "HD Audio" files at 192/24 worth listening to? As distribution, they add nothing audible over well-mastered 44.1/16. As production source material, yes.
Why does my MP3 sound worse after converting from another MP3? Generational loss: the first MP3 discarded information; the second encoder analyzes and discards again over damaged material, amplifying artifacts. Always convert from the best source available.
Is dithering needed when going from 24 to 16 bits? Yes, for music with quiet passages: it replaces correlated truncation noise (metallic-sounding) with harmless low-level white noise.
Convert between MP3, WAV, FLAC, OGG and more with the online audio converter, free and 100% in your browser.