Compress Audio Online

Yes, reducing the bitrate in lossy compression formats (MP3, AAC, OGG) means discarding auditory information. Modern codecs like AAC-LC, developed by the MPEG coalition in 1997 with contributions from Dolby, Fraunhofer, AT&T, Sony, and Nokia, use psychoacoustic models to selectively discard information that the human ear is least capable of perceiving. The psychoacoustic model analyzes the frequency spectrum in time windows (typically 20–50 ms) and identifies two phenomena: frequency masking (a loud sound at one frequency makes weak sounds at nearby frequencies imperceptible, a principle discovered by Harvey Fletcher in the 1920s and formalized in the Fletcher-Munson equal-loudness contours, 1933) and temporal masking (a loud sound makes sounds in the ~100 ms before and ~200 ms after imperceptible). At 128 kbps, AAC and MP3 preserve all perceptually relevant information for most listeners under normal listening conditions. Below 96 kbps, artifacts like pre-ringing, pumping, and high-frequency distortion begin to be audible in complex musical content.

The optimal bitrate choice depends on content and listening context. For podcasts and spoken voice content: 64 kbps mono is the industry standard (used by most podcasts on Spotify, Apple Podcasts, and Overcast), equivalent to approximately 28 MB per hour. At 64 kbps, spoken voice reproduces with complete clarity since the spectral bandwidth of the human voice (85 Hz to 8 kHz for most intelligibility) is significantly narrower than that of music. For casual music streaming: 128 kbps is acceptable for listening on Bluetooth speakers or mid-range headphones. Spotify uses 128 kbps OGG/Vorbis for free users and 160 kbps for premium on mobile. For music with attentive listening: 192–256 kbps. Apple Music uses 256 kbps AAC-LC as the base format for its entire library. For reference files or archiving: 320 kbps MP3 or, better yet, FLAC (lossless). Professional music producers distribute masters in WAV 24-bit/96 kHz to platforms, which then transcode them internally.

The exact formula for calculating compressed audio file size is: Size (MB) = (Bitrate in kbps × Duration in seconds) / 8 / 1024. Concrete examples: a 4-minute song (240 seconds) at 320 kbps occupies (320 × 240) / 8 / 1024 = 9.375 MB. The same song at 128 kbps occupies 3.75 MB (60% less). At 64 kbps it occupies 1.875 MB (80% less than 320 kbps). For comparison, the uncompressed WAV of the same song at 44.1 kHz, 16-bit, stereo occupies approximately 44,100 × 2 × 2 × 240 / 1,024 / 1,024 = 40.6 MB. A 12-track album of 4-minute songs at 128 kbps occupies 45 MB total, versus 487 MB in uncompressed WAV. This factor-10x difference explains why lossy formats were essential to make digital music distribution viable in the 1990s–2000s with the internet speeds of the era (56k modem: 7 KB/s; ADSL 1 Mbps: 125 KB/s).

Lossy compression is a data compression method that reduces file size by permanently discarding information considered non-essential to the user's perceptual experience. In audio, lossy codecs like MP3 (patented by Fraunhofer IIS and Thomson Consumer Electronics, fundamental patents expired in 2017), AAC (ISO/IEC 13818-7:1997 standard), and OGG Vorbis (open source, developed by the Xiph.Org Foundation since 1998) exploit the limitations of the human auditory system to discard information that cannot be perceived. Lossless compression like FLAC (Free Lossless Audio Codec, created by Josh Coalson in 2001) applies algorithms similar to ZIP but optimized for audio signals, typically achieving compression ratios of 2:1 to 3:1 without discarding any information. Lossy compression, on the other hand, can achieve ratios of 10:1 to 20:1 because it discards auditorily irrelevant information, at the cost of the compression being irreversible: the original audio cannot be recovered from the compressed file.

CBR (Constant Bitrate) and VBR (Variable Bitrate) are two bit allocation strategies in lossy audio encoding. CBR maintains a fixed bitrate throughout the entire file: if you choose 128 kbps CBR, every second of audio occupies exactly 128,000 bits, regardless of whether that second contains silence, simple speech, or complex music with high spectral density. This guarantees a predictable file size and facilitates streaming (the server can calculate exactly how many bytes are needed for any position in the file), but is inefficient because it assigns the same bits to simple and complex moments. VBR assigns more bits to the most complex audio segments (higher spectral density, more information to encode) and fewer bits to simpler moments (silences, monotone speech), maintaining more uniform perceptual quality. In MP3, LAME VBR with -V 2 (approximately equivalent to 190 kbps average) produces results indistinguishable from 320 kbps CBR for most listeners in blind ABX tests, with a significantly smaller file size. AAC VBR with quality 3–4 (in FFmpeg) is the equivalent for AAC.

Yes, compressing an already lossy-compressed file (like MP3 to MP3, or MP3 to AAC) implies additional quality degradation compared to compressing from a lossless source (WAV or FLAC). This phenomenon is known as generational loss. Each lossy compression cycle introduces new psychoacoustic artifacts: the artifacts from the first compression (pre-ringing, high-frequency distortion, dynamic range pumping) combine with the artifacts of the second compression, and the psychoacoustic model of the second codec may allocate bits suboptimally because the input material no longer has the statistical properties of natural PCM audio. In practice, at bitrates of 128 kbps or higher, the difference between compressing from WAV and from a 320 kbps MP3 is only audible in highly controlled ABX tests and is imperceptible in casual listening. However, compressing a 64 kbps MP3 to 32 kbps produces clearly audible artifacts even for untrained listeners. The general recommendation is always to start from the highest quality source available for any transcoding operation.