Changing video playback speed involves two distinct technical processes: temporal resampling of the video stream and time-stretching of the audio stream. Understanding both is key to producing professional-quality speed-adjusted content rather than the jarring, distorted output that naive implementations produce.

For the video stream, speed adjustment is fundamentally about temporal resampling, the process of changing how many source frames map to each second of output. When speeding up a video by 2x, every other frame can simply be dropped, halving the duration while maintaining the output frame rate. When slowing down to 0.5x, each frame must be displayed for twice as long, or new intermediate frames must be generated through frame interpolation. Simple frame duplication (repeating each frame twice) produces choppy slow motion because the eye perceives the repeated frames as stuttering. Advanced frame interpolation algorithms analyze motion vectors between consecutive frames and synthesize entirely new in-between frames that show the subject at positions it would logically occupy between the original captures. Optical flow estimation is the primary technique used, computing pixel-by-pixel motion fields to warp and blend source frames into convincing intermediate frames. This is how professional slow-motion effects achieve smooth, fluid motion even when the source was not recorded at high frame rates.

Audio speed adjustment presents an entirely different challenge. Simply playing audio samples faster raises the pitch proportionally, making voices sound like chipmunks, while slowing playback lowers the pitch, creating an unnaturally deep tone. Professional speed adjustment requires time-stretching algorithms that change the duration of audio without altering its pitch. The most widely used approach for speech is WSOLA (Waveform Similarity Overlap-Add), which works by segmenting the audio into short overlapping windows, then repositioning these windows along the timeline while maintaining phase coherence. WSOLA finds the optimal overlap points by analyzing waveform similarity, ensuring that when segments are spliced together, the result sounds continuous and natural rather than producing audible clicks or phase artifacts.

For music and complex audio, more sophisticated techniques like phase vocoders operate in the frequency domain, decomposing the signal into its frequency components using the Short-Time Fourier Transform, stretching the time axis while preserving the frequency content, and then resynthesizing the waveform. This approach handles polyphonic content better than time-domain methods but can introduce artifacts called phasiness in transient-heavy material like percussion. Modern implementations often use hybrid approaches that detect transients and switch between time-domain and frequency-domain processing as needed. The quality of the time-stretching algorithm is what separates professional speed adjustment tools from amateur ones, as poor implementations produce metallic, warbling, or stuttering audio that immediately reveals the speed modification.