Sampling Frequency Conversion

User's Guide > Dialog Reference > Sampling Frequency Conversion

This is a dialog displayed in sampling frequency conversion. The default setting si high quality for speech files, you don't need to change parameters except for expert users. The processing speed is not so fast in compensation for the high quality. So, change parameters based on the following explanation when you use a slow machine.

On the other hand, the default setting seems not to be so high quality for music files such as MP3. In this case, making the sidelobe value bigger (more than 80dB) is recommended.

Sampling Frequency
Target sampling frequency. When the greatest common divisor of the sampling frequencies is small such as 48 kHz and 44.1 kHz, the output sampling frequency is sometimes not identical. The tolerance of the output sampling frequency can be specified by Tolerance.
Normalized Frequency
Normalized cut-off frequency of a lowpass filter. 1.0 means the Nyquist frequency (half of sampling frequency). Smaller value makes the wider frequency region cut off. Too big value can cause aliasing noise. I recommend the value from 0.9 to 0.95.
Normalized Transition
Transition frequency width for a lowpass filter in normalized frequency. 1.0 means the Nyquist frequency. The bigger value makes the lowpass filter having more gentle slope. Too big value can cause aliasing noise. The value from 0.05 to 0.1 is recommended.
Sidelobe
Sidelobe height of a lowpass filter. This parameter gives an amount of attenuation higher than the cut-off frequency. A bigger value makes the amount of attenuation bigger and decreases aliasing noise. However, the bigger value sometimes requires more CPU power.
For speech files, I recommend 40 dB for processing speed, 50-60 dB for quality, 80 dB for very high quality.
For music files, these values are not sufficient, over 100 dB value is sometimes required.
Gain
Gain for sampling frequency conversion. The values more than 1.0 and less than 1.0 make the amplitude bigger and smaller, respectively.
This item is introduced in version 0.6.1.
Tolerance
Tolerance of sampling frequency conversion. A smaller value makes a smaller error. The converted frequency includes an error within the tolerance. A bigger value makes the processing speed fast when the greatest common divisor of the sampling frequencies is small such as 48 kHz and 44.1 kHz. When the greatest common divisor is small, the tolerance is small, and the buffer length is long, very large memory will be required.
When the greatest common divisor is big enough such as 48 kHz and 8 kHz, the tolerance doesn't take effect.
Buffer Length
Buffer length for the processing. Although big value sometimes makes the processing faster, the usage of memory will be bigger.

User's Guide > Dialog Reference > Sampling Frequency Conversion


Last modified: "2004-07-24 21:21:23 hideki"