Engine rejects small but valid WAV files

[illwieckz]

The smallest WAV file the engine accepts is 90 bytes, 8KHz i16 mono.

But the smallest valid WAV file I could produce is 46 bytes:

• null.wav.zip

But the engine rejects it:

Warn: Unhandled exception (St12length_error): cannot create std::vector larger than max_size()

With some other small sizes it sometime even reports that the file is empty!

That looks bogus.

[illwieckz]

It may require 8 samples without actually checking for them:

	AudioData out { sampleRate, byteDepth, numChannels };
	out.rawSamples.assign( audioFile.data() + dataOffset + 8, audioFile.data() + size );

[illwieckz]

It may require 8 samples without actually checking for them

No that's not that, it may be worse, it looks like it's reading from… audioFile.data() + dataOffset + 8 to audioFile.data() + size, so as soon as the size of the samples is smaller than the whole WAV header, the end position is before the start position… It would also mean we always truncated our WAV files.

[illwieckz]

If I'm right, here,

• audioFile.data() is the beginning of the WAV file (first byte of the header);
• audioFile.data() + dataOffset is the beginning of the DATA field in the header, which is 4 byte.
• audioFile.data() + dataOffset + 8 is the beginning of the actual audio data, the address of the samples, after the 4 bytes of the DATA field and the 4 bytes of the size field.

Here is houw looks my smallest null.wav with only one sample:

          file              chunk    format   format channl sample   some            bits per       data    audio
          size              marker   length   type   count  rate     precomputes     sample         size    data
          ↓                 ↓        ↓        ↓      ↓      ↓        ↓        ↓      ↓              ↓       ↓
'RIFF' |  BBBB  | 'WAVE' | 'fmt ' |  BBBB  |  BB  |  BB  |  BBBB  |  BBBB  |  BB  |  BB  | 'data' | BBBB |  BBBB BBBB  |
          ↑                    ↑     ↑        ↑      ↑      ↑                        ↑              ↑       ↑
          byte              0x20     16       1      1      8000                     16             2       1st sample

But if I'm right, the engines attempts to read from the beginning of the sample (8 bytes after the beginning of the data string) to the beginning of the WAVE string, that's all wrong.

It would actually explain why, with 11 samples (that matches what I remember having counted in Audacity), the engine believes the file is empty, because then the end would be the same as the start.

[illwieckz]

Fix in #1853:

• Fix WAV file loading #1853

Now we will be able to embed my 46 bytes sound/null.wav file! 🙂 And we fix a bug at the same time! 😀

constexpr unsigned char sound_null_wav[] =
{
0x52,0x49,0x46,0x46,0x26,0x00,0x00,0x00,0x57,0x41,0x56,0x45,
0x66,0x6d,0x74,0x20,0x10,0x00,0x00,0x00,0x01,0x00,0x01,0x00,
0x40,0x1f,0x00,0x00,0x80,0x3e,0x00,0x00,0x02,0x00,0x10,0x00,
0x64,0x61,0x74,0x61,0x02,0x00,0x00,0x00,0x00,0x00,
};

[illwieckz]

We can actually write it this way:

constexpr unsigned char sound_null_wav[] =
{
0x52,0x49,0x46,0x46, // [RIFF]
0x26,0x00,0x00,0x00, // file size
0x57,0x41,0x56,0x45, // [WAVE]
0x66,0x6d,0x74,0x20, // [fmt ]
0x10,0x00,0x00,0x00, // format length: 16
0x01,0x00,           // format: 1 PCM
0x01,0x00,           // channels: 1
0x40,0x1f,0x00,0x00, // sample rate: 8000
0x80,0x3e,0x00,0x00, // bytes to read per second (sample rate * bytes per block): 16000
0x02,0x00,           // bytes per block (channels * bits per sample / 8): 2
0x10,0x00,           // bits per samples: 16
0x64,0x61,0x74,0x61, // [data]
0x02,0x00,0x00,0x00, // data size: 2
0x00,0x00,           // one sample
};

or this way:

constexpr unsigned char sound_null_wav[] =
{
'R' ,'I' ,'F' ,'F' ,
0x26,0x00,0x00,0x00, // file size
'W' ,'A' ,'V' ,'E' ,
'f' ,'m' ,'t' ,' ',
0x10,0x00,0x00,0x00, // format length: 16
0x01,0x00,           // format: 1 PCM
0x01,0x00,           // channels: 1
0x40,0x1f,0x00,0x00, // sample rate: 8000
0x80,0x3e,0x00,0x00, // bytes to read per second (sample rate * bytes per block): 16000
0x02,0x00,           // bytes per block (channels * bits per sample / 8): 2
0x10,0x00,           // bits per samples: 16
'd' ,'a' ,'t' ,'a',
0x02,0x00,0x00,0x00, // data size: 2
0x00,0x00,           // one sample
};