Quantisation Noise
Definition
The distortion introduced when a continuous audio signal is rounded to discrete digital values. Lossy codecs concentrate quantisation noise at predictable frequency bands determined by the codec's psychoacoustic model and the target bitrate.
- Cause
- Rounding a continuous signal to discrete values
- Distribution
- Concentrated at bands set by codec's psychoacoustic model
- Depends on
- Codec and target bitrate
- Field
- Audio forensics, codec identification
Common questions
How does quantisation noise help identify which codec compressed an audio recording?+
Different codecs allocate bits across frequency bands according to their own psychoacoustic model, so the resulting quantisation noise pattern in a decoded waveform carries a signature analysts can compare against known codec profiles to infer prior compression history.
Does quantisation noise disappear once a lossy audio file is converted back to an uncompressed format?+
No. Converting a lossy file such as an MP3 to WAV changes the container and encoding but cannot restore the information already discarded during quantisation, so the noise pattern introduced by the original lossy encoding remains embedded in the audio.
Related terms
- Double Compression
- The condition in which a video has been encoded at least twice through a lossy codec. The term is also used loosely...
- MEMS Microphone
- A Micro-Electro-Mechanical Systems microphone: the capsule type used in virtually all modern smartphones and tablets. MEMS microphones have a characteristically flat midrange...
- Psychoacoustic Masking Model
- The perceptual model used by lossy codecs such as MP3 and AAC to decide which frequency components can be discarded or coarsely...
- Self-Noise (Noise Floor)
- The electrical noise inherent in a microphone circuit and capsule, measured in dB(A) equivalent input noise. Different microphone types have characteristic noise...
- Spectral Fingerprint
- The characteristic frequency response profile of a recording device, including roll-off curves, resonance peaks, and self-noise level. Comparing a recording's spectral fingerprint...