wav files

Wav files are commonly present in today's computers. A header of 44 bytes and a variable number of data form them. Our goal in this section is to create and read `wav' files.

We are going to create a wav file from a sinusoid at 500 Hz sampled at 22050 Hz with a duration of one second. To calculate the time, we sample the interval from 0 to 1 with a period of 1/22050. After this, we create the sinusoid and save it with the command `savewave' that adjusts all the parameters of the format. We indicate the sampling frequency in the save process

-->fsampling = 22050;                   // sampling frequency
 
-->t = 0 : 1/fsampling : 1;             // one second at this frequency
 
-->sinwav = sin (2 * %pi * 500 * t);    // a wave at 500 Hz
 
-->savewave ('beep.wav',sinwav,fsampling);     // saves as beep.wav
Writing Wave file: Microsoft PCM format, 1 channel, 22050 samp/sec
        44100 byte/sec, 2 block align, 16 bits/samp
Finished writing Wave file, 44102 data bytes

Once wav file has been created, we can hear it with any media player. Now we are going to recover the wave. We could do that with the command `loadwave', but we are going to load the wave from the file directly to show how different types can be loaded.

Wav files are formed by a header part and a data part. As the header part includes binary data with different characteristics, it can be appropriate to read data of different types. The header of a wav file is composed of the following fields

Field	bytes	format	contains
1	0...3	str4	"RIFF" in ASCII
2	4...7	int4	Total bytes minus 8
3	8...15	str4	"WAVEfmt" Eigth character is a space
4	16...19	int4	16 for PCM format
5	20...21	int2	1 for PCM format
6	22...23	int2	channels
7	24...27	int4	sampling frequency
8	28...31	int4	bytes per second
9	32...33	int2	bytes by capture
10	34...35	int2	bits per sample
11	36:39	str4	"data"
12	40:43	int4	bytes in data

With this information we can recover the data. We are going to recover the sin wave that we have just created.

fid = mopen('beep.wav', 'rb');        // opens file to read in binary
ID = mget (4, 'c', fid);              // RIFF
ID = ascii (ID);                      // conversion to ASCII format
Size = mget (1, 'ui', fid);           // total bytes minus 8
typ = mget (8, 'c', fid);             // WAVEfmt_
typ = ascii (typ);                    // conversion to ASCII format
PCM = mget (1, 'ui', fid);            // 16 
PCM2 = mget (1, 'us', fid);           // 1
nchannels = mget (1, 'us', fid);      // number of channels
fsampling2 = mget (1, 'ui', fid);     // sampling frequency
nbbytes = mget ( 1, 'ui', fid);       // bytes per second
nbytescap  = mget ( 1, 'us', fid);    // bytes by capture
nbytessamp = mget ( 1, 'us', fid);    // bytes per sample
worddata = mget (4, 'c', fid );       // data in ASCII format
worddata = ascii (worddata);          // conversion to ASCII format
bytesindata = mget (1, 'ul', fid);    // bytes in data

With this information, we can read the file

wave = mget ( bytesindata / nbytescap , 's', fid); //reads data
mclose(fid)

We can plot the curve and see that everything was OK.

The source code we can see in `macros/sound/wavread.sci' that the way Scilab does this is much safer (it checks inputs and the internal structure of the file). Our goal was not to build a robust routine but to understand how we can read different values from a file that includes binary and ASCII data and that contains in the header the information of the structure utilized to store data.

But what can a clinical neurophysiologist do with wav files?

Besides learning to read and write different types of data, there are several things that could be useful:

• It is possible to hear signals: Almost any computer has a program to hear wav files. In electromyography, hearing signals is important because it is the best way to establish the rhythmical or arrythmical character of intermittent transients immersed in noise and this fact has implications to distinguish fibrillation from end-plate activity. Motor unit potentials have characteristic frequency components that can be isolated by hearing the signal too.
• Being wav format almost universal, there is a lot of tools for conversion to other formats as well as to visualize the signal.
• There is a very active investigation in sound processing with tools that can be directly applied to this format. Among them, compression of data could be useful to clinical neurophysiology.
• Even for non audible signals (for instance, an EEG or a respirogram), we can modify sampling in wave file stored and we can hear great amounts of information in short time. The audition of EEG has been used for the detection of epileptic seizures (as a matter of fact, there are characteristic patterns of frequency displacement associated with epileptic seizures).

We are going to treat the last point with an example. If we sample a respirogram (a signal that follows the respiration by monitoring the nasal flow), the signal is similar to a sinusoid whose frequency is more or less 15/minute, which means that its period is 4 seconds and, consequently, its frequency is 1/4. We can sample it at a frequency of 10 samples per second, which evidently is not an audible frequency. At this rate, 600 samples constitute one minute of signal and 36000, one hour. But we can save data as if they would had been sampled 1000 times faster, at 10000 Hz

-->t = 0: (1/10) : (8*60*60);        // eighth hours at 10 samples per second

-->resp= sin ( 2 * %pi * (1/4) * t); // respirogram at a frequency of 1/4
 
-->savewave('beep2.wav',resp,10000); // saved as if it had been sampled at 10000
Writing Wave file: Microsoft PCM format, 1 channel, 10000 samp/s
        20000 byte/sec, 2 block align, 16 bits/samp
Finished writing Wave file, 576002 data bytes

One hour of recording (36000 data) can be heard in 3.6 seconds and eight hours of recording can be heard in about 30 seconds. With a monotonous respiration at 15/minute we would hear a monotonous tone of 250 Hz. On the contrary, a modification of the amplitude, as can be seen in apneas or hypopneas, can be heard in a very quick way.

We are habituated to "see" neurophysiological signals and the eye is an organ specialized in parallel processing of information, but we can hear neurophysiological signals too and the ability of audition to discriminate signal from noise or to distinguish the frequency components can be useful too.