BitrateCraft · Reference and desk · bitratecraft.org

Transcoding Glossary

This glossary defines the vocabulary used across the rest of the site. Entries are ordered alphabetically and written in the terms an operator meets in a panel or an argument list, so that HandBrake labels, FFmpeg flags and Shutter Encoder fields can be read against a single set of definitions. Where a term has a near-synonym, the synonym is named inside the definition.

Terms A to C

Anamorphic: A picture whose stored pixel grid is not square, so the display aspect ratio differs from the coded width divided by the coded height. Players stretch the image at display time. Ignoring the pixel aspect ratio produces a visibly distorted result even though the frame size is unchanged.
Aspect ratio: The proportion of a picture, written as width to height. The display aspect ratio is what the viewer sees; the storage aspect ratio is what the frame actually contains. Scaling must preserve one of them deliberately, never by accident.
Bit depth: The number of bits used per colour sample. Eight bits per channel is the common delivery format; ten and twelve bits per channel carry finer gradients and are used in intermediate and archival work. Reducing depth is lossy and can introduce banding in smooth areas.
Bitrate: The amount of data consumed per second of playback. Constant bitrate holds a fixed rate, variable bitrate follows scene complexity, and average bitrate aims at a target over the whole file. Bitrate is a consequence of the quality target, not an independent quality setting.
B-frame: A frame coded from motion prediction against frames that come before and after it in display order. Bidirectional frames are cheap in bits but delay decoding, since the decoder must hold references. Too many of them increase latency in live work.
Chroma subsampling: The practice of storing colour information at lower resolution than brightness. Notation such as 4:2:0 states the sampling pattern. Subsampling saves a large fraction of the data with little visible cost, but it complicates colour-keying and precise compositing.
Codec: The algorithm that compresses and reconstructs a stream, short for coder-decoder. A codec defines the bitstream; a container carries it. Choosing a codec and choosing a container are two separate decisions.
Colour space: The mathematical model in which colour is expressed, together with the transfer characteristics that map it to light. Mismatched colour space between source and output is a common cause of a washed-out or oversaturated result.
Constant quality (CQ, CRF): A rate-control mode that targets a fixed level of error instead of a fixed size. The numeric scale is codec-specific: lower values keep more detail. Constant quality is the default choice when file size is allowed to float.
Container: The file format that holds streams, timestamps, track metadata and an index. MP4 and Matroska are the common choices. Changing only the container without re-encoding is a remux.

Terms D to M

Deinterlacing: Reconstructing full frames from interlaced fields. A correct deinterlacer adapts to motion, weaving still areas and interpolating moving ones. Applying it to progressive material damages detail, so it must be enabled only when needed.
Entropy coding: The final lossless stage of compression, which packs symbols into a bitstream. Context-adaptive variants reach slightly smaller files at a small cost in decode time. It never changes the picture.
Frame rate: The number of frames displayed per second. A variable frame rate follows the source's real timing; a constant frame rate imposes a fixed grid. Delivery formats normally expect a constant rate, which is why a conversion may need a rate conversion step.
GOP (group of pictures): A sequence beginning with an independent frame and followed by frames that depend on it. Short groups allow fast seeking and editing; long groups compress better. Group length is one of the strongest influences on both size and seek behaviour.
Hardware acceleration: Encoding or decoding on a fixed-function block rather than the main processor. It is fast and power-efficient, with a smaller search space than software encoding, so quality per bit is usually lower at the same target.
Interlacing: Storing each frame as two half-height fields captured at different moments. It halved the bandwidth of early video but complicates every later processing step, since filters must handle fields rather than whole frames.
Keyframe: A frame that can be decoded without reference to any other. Seeking lands on a keyframe, so their spacing bounds seek precision. Also called an intra frame or I-frame.
Loudness normalisation: Adjusting audio gain to a measured loudness target rather than a peak target. It makes programmes consistent between each other without clipping. Peak normalisation alone leaves perceived volume uneven.
Mixdown: Rebuilding audio into a different channel layout, for example folding surround into stereo. A mixdown decodes and re-encodes the audio, unlike a pass-through, and may apply downmix coefficients.
Muxing: Writing streams into a container, including timestamps and track headers. Muxing without re-encoding preserves the coded data exactly. Remuxing is muxing a stream that already exists.

Terms P to V

Pixel format: The description of how samples are arranged in memory: bit depth, colour model and subsampling together. Both sides of a filter chain must agree, or the tool inserts a conversion that costs time and can shift colour.
Preset: A named bundle of encoder settings. Presets in HandBrake describe a delivery target; encoder presets inside a codec describe a speed-versus-efficiency trade. The two meanings should not be confused.
Profile and level: Declared constraints that tell a decoder what it must support, covering features and maximum rates. A stream that exceeds the level of a playback device will not play, however well it was encoded.
Quantisation: The step where transform coefficients are reduced in precision. It is the main lossy stage and the direct object of the quality scale. Higher quantisation means fewer bits and more visible error.
Remux: Rewriting streams into another container without touching the coded data. It is fast and lossless, and it is the correct answer when only the wrapper is wrong.
Sample rate: The number of audio samples per second. Resampling is cheap and mostly inaudible, but it is a conversion, so a batch should use one rate throughout to avoid repeated passes.
Telecine: The process that maps film frames onto video fields, historically by repeating fields. Reversing it must detect the repetition pattern; forcing a frame rate without detection produces judder.
Transcode: Decoding a stream and encoding it again, usually with different codec, resolution or quality settings. It is lossy for lossy sources, so it is applied once and then left alone.
Variable frame rate: A stream whose frame intervals vary with the source timing. It is efficient for screen recordings, but it complicates synchronisation, and many delivery formats expect a constant rate instead.
Vectors and motion estimation: The search for the best match between a block and its neighbours in other frames. Search depth is the main dial behind a speed preset: wider search, better match, smaller file, more time.