Audio samples are processed with accurate and ultra-precise (64-bit floating point) algorithms.
Audio Engine uses professional quality SoX resampler (licensed under LGPL). See information about the SoX resampler at SoX - Sound eXchange | Resampling.
SoX resampler is the default sample rate converter in all conversions. SoX resampler is also used with the Opus, MP3, and DSD encoders. Opus and MP3 usually use their own internal sample rate converter. SoX sample rate converter is used to achieve even more precise, better audio quality.
Portions of the information below is quoted from the official SoX website.
Resampler's passband setting determines how much of the frequency content of the original signal (wrt the original sample rate when up-sampling, or the new sample rate when down-sampling) is preserved during conversion. The term 'passband' is used to refer to all frequencies up to the bandwidth point (e.g. for 44.1kHz sampling rate, and a resampling passband of 95%, the passband represents frequencies from 0Hz to ~21kHz). Increasing the resampler's passband results in a slower conversion and can increase transient echo artefacts.
Default sweep 95% - Conversion from 96kHz to 44.1kHz
Default passband 95% - Conversion from 96kHz to 44.1kHz
Steep sweep 99% - Conversion from 96kHz to 44.1kHz
Steep passband 99% - Conversion from 96kHz to 44.1kHz
Aliasing above the passband is allowed. For example, with 44.1kHz sampling rate, and a resampling passband of 95%, this means that frequency content above ~21kHz can be distorted; however, since this is above the passband (i.e. above the highest frequency of interest/audibility), this may not be a problem. The benefits of allowing aliasing are reduced processing time, and reduced (by almost half) transient echo artefacts.
Aliasing is automatically disabled if the source or destination sample rate is below 44.1kHz
Default sweep 95%, aliasing allowed - Conversion from 96kHz to 44.1kHz
Default passband 95%, aliasing allowed - Conversion from 96kHz to 44.1kHz
Steep sweep 99%, aliasing allowed - Conversion from 96kHz to 44.1kHz
Steep passband 99%, aliasing allowed - Conversion from 96kHz to 44.1kHz
When the bit depth is reduced (quantized) there will be quantization errors. I.e. lower bit depth can't accurately represent the source signal. When each sample value is rounded to the nearest value, then the steppiness in the audio may be audible, especially with very quiet sounds. Dither (intentionally applied form of noise) can be applied to avoid it.
Dither can be applied (when enabled from the Settings) if converting from original source of
· 64-bit, 64-bit float, 32-bit, 32-bit float to 24-bit
· 64-bit, 64-bit float, 32-bit, 32-bit float, 24-bit to 16-bit
· 64-bit, 64-bit float, 32-bit, 32-bit float, 24-bit, 16-bit to 8-bit
Noise-shaped dither has higher peak amplitude than other forms of dither, but the noise is concentrated in the very high frequency range where it is less audible. Noise-shaped dither implements Lipshitz minimally audible noise shaping.
Triangular dither produces a lower peak amplitude of noise than noise-shaped, but the noise is concentrated less in the higher frequencies.
Rectangular dither produces a similar noise amplitude as triangular, but with very small amounts of randomization. The frequency distribution is fairly even, so that it approximates white noise. Unlike the other two forms of dither, rectangular does not add noise to absolute silence.
Round samples to nearest value (better audio quality than alternative of truncating samples)
DSD to PCM conversion decimation low pass filter quality setting. Both filters provide flat response up to 30 kHz [DSD64], 50 kHz [DSD128], 100 kHz [DSD256].
Faster, looser decimation low pass filter. Noise level below -176 dB.
Slower, more accurate, steeper decimation low pass filter. Noise level below -191 dB.
DSD64 to PCM noise levels