Differences in Encoding/Decoding Stereo Streams and AIFF and WAVE Files
Apple's and Microsoft's ADPCM compressed files also differ in how stereo streams are encoded, and therefore decoded. The IMA standard does not say anything about having predictor bytes, nor does it say how multiple channels of sound should be treated, and that is one place where AIFF and WAVE files differ.
For packet size, Apple chose 64 samples per predictor bytes, and for stereo, Apple defines that a packet of left channel data precedes a packet of right channel data. Apple's IMA compressor puts two bytes at the front of each packet, which are referred to as predictor bytes (not be confused with the predictor value that the IMA-ADPCM algorithm defines). These two bytes contain the step index, a 7-bit value in the low byte, and the most significant 9 bits of the predictor value. Obviously, since the real predictor value is 16 bits, this 9-bit predictor is only an approximation, but it doesn't cause significant loss of quality.
Microsoft, on the other hand, lets the program compressing the sound determine the size of the packet, and the WAVE standard is eight samples of left data followed by eight samples of right data. WAVE files also have a two byte header just like the Mac, but have an 8-bit step index instead of a 7-bit step index and an 8-bit predicted sample approximation instead of a 9-bit approximation.
For more information on the WAVE format, interested developers may want to consult one or more of the specifications and descriptions that can be found on the Internet
The wBlockAlign Field
The WAVE file format has a field called wBlockAlign which tells the program reading the file the minimum number of bytes that must be processed at a time. The only WAVE files compressed with ADPCM that I have studied have had this value set to either 512 or 1024. This field could be set to other values. Setting this value to a number larger than 34 allows the WAVE file to have fewer predictor bytes than the same data as an IMA compressed AIFF file.
Apple's method of small packet sizes gains the ability to start playing sound from more locations in the file, thus giving better random access. Microsoft's method still allows for random access, but with greater jumps between points where sound can begin playing from.
Back to top
Consequences of the Differences Between AIFF and WAVE File Formats
Because of the substantial differences between the AIFF and WAVE formats, the Apple IMA decompression codec will not handle Microsoft's ADPCM compressed WAVE files.
Stereo ADPCM WAVE files have the channels interleaved differently than AIFF files and have fewer predictor bytes, which is a significant enough difference that the Apple IMA decompressor will not decompress them and produce the same sound as the Microsoft ADPCM decompressor would. The Microsoft ADPCM decompressor will not decompress a stereo IMA AIFF file as the Apple IMA decompressor would unless it is modified to take into account the different channel interleaving and predictor bytes.
Because the Apple IMA compressor makes use of the high bit of the low byte in the predictor word for the predicted sample (to double its precision) and the Microsoft IMA compressor treats this bit as part of the value of the step index, it is likely that the step index will be invalid most of the time if the Microsoft ADPCM decompressor is dealing with an Apple IMA compressed sound.
The step index is likely to be invalid because, depending on what the value of the predicted sample is, the high bit of the low byte of the predictor word (the least significant bit of the predicted sample) may or may not be set. Because the Microsoft ADPCM decompressor takes all 8 bits of the low byte of the predictor word, instead of only the low 7 bits (which are all that are needed; the step index's possible range is from 0 to 88, inclusive), having the high bit set will automatically make the index out of range, which may cause the decompressor to stop decompressing data.
Differences in Resulting Data at the Nibble Level
Inspection of the same data compressed using Apple's IMA compressor and Microsoft's ADPCM compressor shows differences in the resulting data at the nibble level. The difference in the resulting data is often not major: if the data is different, it is usually only off by one value plus or minus. Examining the public Microsoft IMAADPCM.C file shows that the imaadpcmFastEncode and imaadpcmSampleEncode algorithms are not exactly the same as that published by IMA, but I have not attempted to trace execution of sample data to determine where the algorithm is producing different data. When played back through the correct decompressor, the two sounds sound identical.
The Bottom Line
What this means is that if you were to go to all the work of converting from one file format to the other -- for example, converting a WAVE file to AIFF file -- you would find that when you decompressed the sound, the resulting sound would be slightly different, though the change might not be audible.
in the later years as my interest in hiphop has gone beyond dr dre, snoop dogg and 2pac, i'm glad i did though, because a lot of those albums have some quality tracks on them. if you rip your tupac tracks in wav format your 250gb harddrive is going to fill up pretty soon. each track will probably average at 120mb, which is a lot. that's what you have the source item for. take better care of your cds, tapes and vinyl instead. you'll find tons of different ways to take care of them - for instance there are special covers for vinyls that preserve them better and that keeps dust away. i'm sure there are similar solutions for tapes and cds.
