Lossy Audio

有损音频 refers to a data format in which some signal information is permanently discarded or approximated during encoding in exchange for a lower bit rate and smaller file size. Formats such as MP3, AAC, Dolby Digital, DTS, and Opus employ various transformation, prediction, quantization, and entropy coding techniques. Many of these formats take advantage of the human auditory system’s frequency resolution, masking, and temporal characteristics to prioritize errors in positions where they are less likely to be noticed.

Lossy encoding does not mean randomly deleting frequencies. Encoders typically divide audio into temporal or frequency units, estimate the amount of noise each section can tolerate, and then allocate a limited number of bits to the quantized parameters. Insufficient bitrate, improper handling of transients, or repeated transcoding can cause distortions such as pre-echo, metallic sound, changes in soundstage, and harsh high frequencies; The specific manifestations depend on the format, encoder, settings, and source material—not solely on the file extension.

Decoders can generate standard PCM for playback but cannot recover details that were not preserved during encoding. Converting a lossy file to WAVE, FLAC, or a higher bit depth merely preserves the decoded result in a lossless manner; re-encoding it into another lossy format may result in cascading losses. If editing or remastering is necessary, lossy sources are typically first decoded into a high-precision working format to avoid repeated compression at each step, though this still cannot restore the original data.

Fixed bitrate, variable bitrate, and average bitrate describe data allocation strategies. Variable bitrate allocates more data to complex segments and less to simple ones, and typically better reflects quality objectives than a single frame bitrate; the overall average displayed by media tools does not reveal the quantization intensity at every moment. Bitrate figures across different formats cannot be directly compared.

“Transparency” is often used to describe a state in which it is difficult to distinguish the encoded result from the reference signal in specific listening tests; it depends on the listener, equipment, source material, and testing methods, and does not mean that a lossy file has become lossless. A “high bitrate” label, a wide frequency spectrum, or a large file size does not prove that the source has not been encoded at a lower bitrate; determining the source requires a combination of bitstream characteristics, distribution channels, and repeatable comparisons.