Why is audiowrite scaling my int32 data?

I run the following with confusing results:

>> audiowrite('test.wav', -2^32:2^23:2^32, 10000, 'BitsPerSample', 32)
>> dataRead = audioread('test.wav');
>> max(dataRead)

ans =

     4.2950e+09 %%%%%As expected

>> min(dataRead)

ans =

    -4.2950e+09 %%%%%As expected

>> audiowrite('test.wav', int32(-2^32:2^23:2^32), 10000, 'BitsPerSample', 32)
>> dataRead = audioread('test.wav');
>> max(dataRead)

ans =

      1.0000 %%%%%Huh?

>> min(dataRead)

ans =

      -1 %%%%%Huh?

Why does the "int32" data get scaled when it should fit in the range specified by audiowrite help:

    Data Type of Y    Valid Range for Y
    -----------------------------------
        uint8            0 <= Y <= 255
        int16       -32768 <= Y <= +32767
        int32        -2^32 <= Y <= 2^32-1
        single        -1.0 <= Y <= +1.0
        double        -1.0 <= Y <= +1.0

    Data beyond the valid range is clipped.

    If Y is single or double, then audio data in Y should be normalized 
    to values in the range -1.0 and 1.0, inclusive.

    Output Data Type
    The native data type written to the audio file is determined by 
    the file format, the data type of Y, and the specified output 
    BitsPerSample.

    File Formats  Data Type of Y     Output BitsPerSample  Output Data Type
    -----------------------------------------------------------------------
    WAVE (.wav)    uint8,int16,int32         8                  uint8
                  single,double             16                 int16
                                            24                 int32
                  ---------------------------------------------------------
                  uint8,int16,int32         32                 int32
                  ---------------------------------------------------------
                  single,double             32                 single
                  ---------------------------------------------------------
                  single,double             64                 double

[Mr. Mas]

Thanks, Geoff. That's the information I needed.

Answer 1

[Geoff Hayes]

Mason - I don't think that the problem is with audiowrite for your int32 example. According to audioread, If you do not specify dataType, or dataType is 'double', then y is of type double, and matrix elements are normalized values between −1.0 and 1.0.. So

 >> audiowrite('test.wav', int32(-2^32:2^23:2^32), 10000, 'BitsPerSample', 32)
 >> dataRead = audioread('test.wav');

will be such that dataRead is an array of doubles normalized between -1.0 and 1.0. If I do the following instead

 >> dataRead = audioread('test.wav', 'native');
 >> whos dataRead

 Name             Size            Bytes  Class    Attributes

 dataRead      1025x1              4100  int32

then dataRead is of the correct data type with the minimum and maximum the expected integer values.

The problem seems more to be with

 audiowrite('test.wav', -2^32:2^23:2^32, 10000, 'BitsPerSample', 32)

where the input -2^32:2^23:2^32 is of a single (or double) data type. According to the documentation for audiowrite, I would have expected the data to be clipped to [-1.0, 1.0] but that doesn't seem to be the case. With

 >> audiowrite('test.wav', -2^32:2^23:2^32, 10000, 'BitsPerSample', 32)
 >> dataRead = audioread('test.wav');
 >> whos dataRead

Name             Size            Bytes  Class     Attributes

dataRead      1025x1              8200  double

with the minimum and maximum of -4.2950e+09 and 4.2950e+09 respectively. This seems unexpected - why wasn't this clipped as the documentation suggests it would be?

[Mr. Mas]

Funny how I misunderstood what to expect exactly the opposite of what happened. No idea about the clipping -- but I can see now, plain as day, that it should have been clipped, not the values I said were expected.