An audio signal consists of a sequence of samples. The sample is a value of a digital signal at the moment. They follow one-to-one with an equal time frame.
Sample rate is a sample (frame) number per second. A higher sample rate means more samples per second.
An audio sample rate converter alters the time interval between samples of a digital signal. The signal is not changed. The time frame is changed. Samples are recalculated to keep unchanged content of the digital signal.
The sample rate may be changed to adapt audio recording for abilities of certain music device. In instance, your can adjust sampling rate for your digital-to-analog converter.
Sample rate conversion may be done in real time ("audio stream on fly") or offline: converting of audio resolution of files before playback.
In instance, computer's operating system normalizes discretization frequency to one value from different sound streams in real time. It allows using one DAC for playback of the stream mix.
For audio equipment, 2 standard ranges of sampling frequencies are accepted:
Resampling may causes some quality losses. And the best sample rate converter should keep the quality on the highest level. In ideal, the losses should not be audible for user.
Read how resampling impacts sound quality...
To convert/change sample rate:
Remark: Free version has some limitations.
The algorithm for sampling frequency change consists of following steps:
1) Increase up of sampling rate (oversampling) to frequency, multiply target signal's sampling frequency.
2) Filtration of "parasitic" signals (named "artifacts") above half of the target frequency of digitization.
3) Make multiple decimation of superfluous samples for decreasing sampling rate (downsampling).
An increase of sample frequency is made by way of inserting additional ("virtual") samples between "real" samples of input digital signal. These "virtual" samples are generated by the interpolator.
The insertion of "virtual" samples, containing zero-level values, may be applied. It is a faster method for calculations than spline function, in instance. The last oversampling way adds a significant amount of "artifacts".
Why it is necessary to increase frequency of digitization? It's needed to execute goal 3) of the algorithm. Sample decimation is most easier when it is multiple: just delete superfluous.
Further "parasitic" signals, with frequencies above half of target sampling rate, are filtered. Otherwise, after deleting "superfluous" samples, "artifacts" will added to the spectrum of an audible signal. They will distort it, or add noise. "Bad" sound will happen.
For reduction of distortions, added to converted signal, we should interpolate it as much as possible precisely. Interpolation accuracy is closest coincidence of initial analog signal with interpolator's "virtual" samples. It is necessary to remember, that the top-quality interpolator can restore an initial analog signal precisely enough. But not with 100 % accuracy. Alas. At increase of sampling frequency there will be "parasitic" signals above half of output signal's frequency of digitization.
During development of hi-end sampling rate converters, special attention is paid to quality of low frequencies filter. If this filter will not suppress "artifacts", it will mixed with useful signal after deleting of "superfluous" samples.
For demonstration of filtration's quality we shall look at the time-diagram of spectrum. Horizontal axis there is time, vertical axis - frequency. Vertical lines - is spectrum into time point at horizontal axis. The level of signal is shown by color (white - the highest, black - the lowest - less minus 150 dB). Sinusoidal audio signal with sweep frequency (height of tone) come to input of sample rate converter.
Here such result will be on an output of hi-end audio converter:
We see only recurrence of input signal, without in addition appearing frequency components - "artifacts". Level of artifacts less minus 150 dB. Less quantizations error level for 24-bit samples (minus 144 dB). Inevitable loose quantizations errors will mask "artifacts".
The high and middle quality audio converter of sampling frequency will give a following picture:
Here shown "artifacts" having a level near minus 105-110 dB (dark blue color) . These "artifacts" arise, both at interpolation, and at insufficient suppression (before decreasing sampling rate) of the "parasitic" signals located above half of frequency of digitization.
Let's look for comparison the spectral diagram of too low quality converter:
In this case "artifacts" achieve a level near minus 50 … 60 dB (red and pink colors).
The filter should pass audio a signal at different frequencies from 0 up to 20000 Hz without changing loudness useful. For this purpose inflating of the frequency characteristic (on different frequencies at passage through the filter) should not exceed change of a level of loudness of output signal less 1 … 2 dB.
As much as possible to keep waveform of transformed audio signal it is required to provide an identical time delay for all spectral components at passage through the filter. It is provided, if the filter have linear phase-frequency characteristic. Linear is means in the form of a flat inclined line.
At such form of the phase characteristic all spectral components have an identical delay. The signal passes through the filter not deformed{distorted}. It is especially important for maintenance of quality sounds with sharp attack (drums, piano, guitar, etc.).
Besides filters have such feature, as "ringing". It parasitic filter's output self-oscillation by sharply changing of input signal. Apply impulse to input of converter. At output the impulse turns to the oscillation stretched on time. "Ringing" is heard as click.
The more "sharp" recession to a level between pass bands and suppression band of filter, the more level of "ringing".
Therefore it is necessary for developer of the converter to choose the compromise between a sharpness of recession of filter's amplitude-frequency characteristic (that in the most positive influences to suppression of "artifacts") and level of "ringing". Very heavy case, when final result of conversion is sample rate 44,1 kHz. Between the maximal frequency of a useful signal (20 kHz) and half of frequency of digitization (22,05 kHz) the difference on frequency makes only 2,05 kHz. At a desirable degree of suppression of artifacts nearby 140 dB!!!
Quality of sampling rate conversion depends at:
- quality of interpolation;
- value of oversampling frequency;
- quality of filtration before decreasing of sampling rate.
For the majority of modern records 24-bit samples is used. It achieve theoretical level of quantizations noise minus 144 dB. Accordingly the level of all transformation's "artifacts" should not exceed minus 144 dB. Thus "artifacts" will masked in noise of quantizations. There is no reasons make level of "artifacts" less noise of quantizations.
Quality of converting signal is provided with filter's linearity phase and evenness of frequency response.
See description high precision sample rate converter AuI ConverteR 48x44.
You can convert sample rate according to the manual...
Sampling rate 44.1 kHz is minimum discretization frequency to transmit audible range. Minimum frequency causes additional issues for developers of audio equipment and software.
Read more...
You can change sample rate from 44.1 to 48 kHz according to the guide...
March 23, 2023 updated | since July 3, 2011
