Conversion and quality

[Previous] [Main] [Next]

Converting files from one format to another does not affect sound quality if all that changes is the data-recording format. Compression, companding and sample rate changes of digitized data are subject to the laws of physics and do cause a slight or sometimes drastic alteration in sound quality. First you have the expected degradation resulting from the target sample rate conversion itself. When you down-convert the sample rate of a file you always lose quality because you lose information. These are the laws of physics, and we have to accept them. The resulting quality is not much different than what the file would have been if it were recorded directly at the lower sample rate or resolution. For example, converting a hi-fi 16-bit linear ".wav" file at 44 KHz into a 4-bit ADPCM file at 6 KHz does cause a very noticeable (and unavoidable) degradation in sound quality. This is because a 44 KHz sampled signal can contain components up to a frequency of 22 KHz, while a 6 KHz signal can only contain components up to 3 KHz! In effect you are going from a hi-fi quality signal to a low telephone-quality signal. Also, only 4 bits now represent your original 16-bit signal. Nobody can avoid these basic laws of physics. This is not as dramatic as it sounds, because in the end you are going to use this message over telephone lines anyway and telephone bandwidth is only from 300 to 3,400 Hz. You also should expect a very slight but noticeable degradation resulting, essentially, from the resampling processes themselves. Record directly into the final target sample rate if your hardware allows it and does it well. Always check the sound quality of any conversion process on a typical original before committing hundreds of files to it. Remember that differences are less noticeable when played back over a voice telephony card and to the telephone network. Check quality on the final target hardware.

A clean source file converts much, much better than one with background noise. This may be surprising, but tiny background noises usually end up amplified in the converted file. Do not spend time tweaking Vox Studio, it does not need it. Spend your time making sure your original recordings are of the highest possible studio quality. High quality recordings convert very nicely, but junk remains junk. Try the example recordings that came with Vox Studio to benchmark conversion quality. You will see conversion quality is superb if your original is of comparable quality. Always try to record your originals at 16-bit resolution, not at 8-bit.

Vox Studio can up-convert a low-sample-rate, low-resolution telephony file back to, for example, 44 KHz and 16 bits. The resulting sound quality will be similar to the original, however. You cannot gain bandwidth, and the signal information that was absent in the original will also be absent in the up-converted file. You cannot change a telephone-quality file into a hi-fi-quality file, but you can do the opposite.

Vox Studio uses digital signal processing to perform the signal conversions. Each conversion involves many mathematical manipulations of the recorded files. It is best to avoid unnecessary conversion steps and record or convert your master file directly to the desired target format if your sound card records well at that sample frequency. Many sound cards unfortunately do not have good recording capabilities below 11 KHz sample rates; they were optimized to be used at 11, 22 and 44 KHz and nothing else.

If you plan to edit your files, do this editing on a copy of your master ".wav" recording, before conversion to telephony format. Convert to telephony format as the last step in the process.

All the above may seem complicated but we only discussed it so you would understand the various parameters that can affect the quality of a recorded or converted message. The reality is that, in general, you can obtain excellent telephony quality prompts if you follow a few simple guidelines:

·	Record the originals as 16-bit ".wav" files at 22 KHz or higher.
·	Make sure your original recordings are spotless: good intelligibility, good sonority, no background noise, good volume, etc. If they are not, start again until they are. Do not attempt to convert unless the original recordings are perfect and sound professional.
·	Use Vox Studio to convert to the highest sample rate allowed by your telephony hardware (for instance, 8 KHz sample rates are better than 6 KHz) and use A-law or Mu-law rather than ADPCM coding if you can.