next up previous contents
Next: Format Conversion Up: Post-processing Previous: Filtering   Contents


Re-sampling

Very often you will find the situation that your recording device does not record with the desired sampling rate as specified for the final corpus. A typical case is the recording with a DAT recorder which usually allows only either 48kHz or 44.1kHz sampling rates. These high sampling rates are required for HiFi recordings but not for speech where a maximal sampling rate of 22.05kHz is sufficient.

To save space in the final distribution the signals have to be down-sampled. Prior to down-sampling you have to be sure that the recorded signals do not contain any frequencies higher than half of the intended sampling frequency after down-sampling7.7. You may either take care of that in the recording process itself or filter the raw data using a low pass filter before down-sampling.

There might also be the case that you have to re-sample your data to higher sampling frequencies, for instance to meet special requirements of your partners, an annotation tool or your client. In this case no information is added to the signal and therefore no filtering is necessary.

Re-sampling can be done using public domain tools like sox7.8. Some professional tools automatically filter the signals before re-sampling. Check their respective manual to be sure. Be aware that re-sampling in most cases causes a degradation of quality of the raw signal. It depends on the algorithm used and the sample format of the data how good the quality of re-sampled data will be. In most cases we have found that sox delivers sufficient quality. More importantly, sox explicitly states how rate conversion is performed - most other applications and tools do not disclose this information.


next up previous contents
Next: Format Conversion Up: Post-processing Previous: Filtering   Contents
BITS Projekt-Account 2004-06-01