Hop Size: Frame Shift, Window Overlap and STFT

The sample or time offset between the starting positions of consecutive analysis windows in frame-based signal processing.

Hop size is the amount by which an analysis window advances between consecutive frames.

For 16 kHz audio, a 10 ms hop corresponds to:

16000 × 0.010 = 160 samples

Relationship with Window Length

With a 25 ms window and a 10 ms hop, consecutive windows overlap. A smaller hop increases temporal sampling of the analysis but also increases the number of frames and compute cost.

Time Axis

A simple frame start is:

start_i = i × hop

Padding, centered windows or resampling can shift the effective timeline, so library-specific semantics should be verified when exact timestamps matter.

Speech Processing

MFCC and STFT pipelines use hop size as a basic time-axis parameter. It is different from a VAD segment boundary: the VAD segment is a larger speech region, while hop size is the internal step between analysis frames.

Related: Frame, STFT, Window Function.