Sample-rate conversion: expose aliasing and filtering

Exact evidence inspector

Read the selected state numerically. This includes the details behind the chart rather than a preset verdict.

A new sample grid is not new information

The upper trace is a synthetic sine sampled at the input rate; the lower trace is the converted output over the same twelve-millisecond window. Conversion changes the sample grid, not the physical duration or the information already lost during sampling. Linear interpolation blends the two neighboring input samples. The filtered method evaluates a windowed-sinc kernel over a sixty-four-input-sample span with a Hamming window and normalized weights. Its cutoff is forty-five percent of the smaller sample rate, leaving a transition band below the output Nyquist frequency.

Choose rates, tone and method

Input sample rate and Output sample rate offer eight, sixteen, twenty-four, forty-four-point-one and forty-eight kilohertz. Synthetic tone frequency ranges from one hundred hertz to ten kilohertz, so some choices deliberately exceed the input Nyquist limit. Conversion method selects unfiltered linear interpolation or finite windowed-sinc filtering. Each change recomputes the samples immediately; this is a direct numerical tool, not a timed animation. Select output sample moves the red marker and displays its index, time, amplitude and fractional position on the source grid. Restore defaults returns to a one-kilohertz tone converted from forty-eight to sixteen kilohertz with filtering.

Make an alias visible

Choose an input rate of forty-eight kilohertz, output rate of sixteen kilohertz, tone frequency of ten kilohertz and linear interpolation. The output grid cannot distinguish this tone from a folded six-kilohertz component. Switch to windowed-sinc: output RMS should fall sharply because the tone lies above the anti-alias cutoff. Return to one kilohertz and compare; the wanted low-frequency tone should remain. Now choose eight-kilohertz input with a ten-kilohertz tone. The input has already aliased to two kilohertz. Raising the output rate cannot reconstruct the original ten-kilohertz signal from those samples.

Understand the plotted boundaries

The result reports input and output sample counts, output RMS, the frequency represented on the input grid and its unfiltered output fold. The fold label is a mathematical candidate, not a measured spectral peak; filtering may suppress it. Noninteger sample counts are rounded for the displayed finite window. The synthetic sine is extended beyond both edges when the interpolation kernel needs neighboring samples, avoiding an artificial zero-padding transient; a real stream must buffer samples and handle boundaries explicitly. This demonstration is not a production resampler: filter order, stopband attenuation, phase response, quantization, multi-channel data, streaming delay and computational cost are not benchmarked. The fixed finite filter has a transition band and cannot eliminate every alias. No file is uploaded and no audio is recorded or played.