No compatible ACM codec to decode 0x00FF audio stream to PCM

[StainlessS]

Yes, you only get error when you try to play in VD. VDMod just outputs silence, no error.

VLC plays Ok, as does SMPlayer, UMPLayer, KMPlayer, Pot Player Media Player Classic HC, WMP v9.0, MS WMP v6.4(xp old version player).
Only failing player found was GOM.

[IanB]

It seems Fcchandler's AAC ACM codec does some validation on the stream format structure contents.

On your included sample wfex->cbSize is 0, Fcchandler is expecting a minimum of 2 bytes of codec specific data, apparently that being the two-byte AAC AudioSpecificConfig. But I cannot see that he accesses that information anywhere so it might be possible to hack the code to just ignore it.

I hex edited your sample wfex->cbSize from 0 to 2 and now VDub accepts your sample with a header needs repair warning.

When played it appears mostly silent with some shuffling noise in the background, is this correct.

Do you have another sample with some distinctive sound, i.e. count 1 2 3 to check.

static bool IsValidAAC(const WAVEFORMATEX *wfex, const MyData *md)
{
	// Does "wfex" describe an AAC format we can decode?

	if (wfex != NULL)
	{
		int i;

		if (wfex->wFormatTag != WAVE_FORMAT_RAW_AAC1)
		{
			if (!(md->dwFlags & AACACM_706D)) goto Abort;
			if (wfex->wFormatTag != WAVE_FORMAT_FAAC51) goto Abort;
		}

		for (i = 0; i < AACACM_NCHANS; ++i)
		{
			if (wfex->nChannels == chan_map[i]) break;
		}
		if (i == AACACM_NCHANS) goto Abort;

		for (i = 0; i < AACACM_NFREQS; ++i)
		{
			if (wfex->nSamplesPerSec == freq_map[i]) break;
		}
		if (i == AACACM_NFREQS) goto Abort;

		if (wfex->nAvgBytesPerSec == 0) goto Abort;
		if (wfex->nBlockAlign == 0) goto Abort;
		if (wfex->cbSize < 2) goto Abort;
		
		return true;
	}
Abort:
	return false;
}

static inline unsigned short make_aac_config(DWORD srate, WORD chans)
{
	// Create two-byte AAC AudioSpecificConfig

	unsigned short asc = 0;

	// AudioSpecificConfig, 2 bytes

	//   LSB      MSB
	// xxxxx... ........  object type (00010 = AAC LC)
	// .....xxx x.......  sampling frequency index (0 to 11)
	// ........ .xxxx...  channels configuration (1 to 7)
	// ........ .....x..  frame length flag (0 = 1024, 1 = 960)
	// ........ ......x.  depends on core coder (0)
	// ........ .......x  extensions flag (0)
		
	//   LSB      MSB
	// .....000 0....... = 96000
	// .....000 1....... = 88200
	// .....001 0....... = 64000
	// .....001 1....... = 48000
	// .....010 0....... = 44100
	// .....010 1....... = 32000
	// .....011 0....... = 24000
	// .....011 1....... = 22050
	// .....100 0....... = 16000
	// .....100 1....... = 12000
	// .....101 0....... = 11025
	// .....101 1....... =  8000

	switch (srate) {
	case 96000: break;
	case 88200: asc |= 0x8000; break;
	case 64000: asc |= 0x0001; break;
	case 48000: asc |= 0x8001; break;
	case 44100: asc |= 0x0002; break;
	case 32000: asc |= 0x8002; break;
	case 24000: asc |= 0x0003; break;
	case 22050: asc |= 0x8003; break;
	case 16000: asc |= 0x0004; break;
	case 12000: asc |= 0x8004; break;
	case 11025: asc |= 0x0005; break;
	case  8000: asc |= 0x8005; break;
	default:
		return 0;
	}

	//   LSB      MSB
	// ........ .0001... = 1 channel
	// ........ .0010... = 2 channels
	// ........ .0011... = 3 channels
	// ........ .0100... = 4 channels
	// ........ .0101... = 5 channels
	// ........ .0110... = 6 channels
	// ........ .0111... = 8 channels

	switch (chans) {
	case 1: asc |= 0x0800; break;
	case 2: asc |= 0x1000; break;
	case 3: asc |= 0x1800; break;
	case 4: asc |= 0x2000; break;
	case 5: asc |= 0x2800; break;
	case 6: asc |= 0x3000; break;
	case 8: asc |= 0x3800; break;
	default:
		return 0;
	}

	return (asc | 0x0010);
}

[belonesox]

Originally Posted by IanB

Do you have another sample with some distinctive sound, i.e. count 1 2 3 to check.

This file generated by ffmpeg with "-acodec libvo_aacenc"
There another sample with some distinctive sound.

[StainlessS]

Thread link and problem reported to GOM Labs.

[Wilbert]

On your included sample wfex->cbSize is 0, Fcchandler is expecting a minimum of 2 bytes of codec specific data, apparently that being the two-byte AAC AudioSpecificConfig. But I cannot see that he accesses that information anywhere so it might be possible to hack the code to just ignore it.

Is AudioSpecificConfig mandatory (and the sample not spec compliant)?

Another sample: http://www.wilbertdijkhof.com/avi-aac.zip (contains AAC LC 48 khz 5.1 channels). bytes of AudioSpecificConfig = 11B0. Yes, i think the channels are wrongly coded, but it opens and plays fine (Virtualdub and AviSynth).

Btw, AudioSpecificConfig is larger than 2 bytes in case of HE-AAC with explicit signaling of SBR/PS: http://msdn.microsoft.com/en-us/library/windows/desktop/dd742784%28v=vs.85%29.aspx

[IanB]

Hacking the wfex->cbSize works on the new sample, I hear a guy speaking in Russian.

I will have a proper look at the AAC ACM source code over the weekend see what the downside of removing the wfex->cbSize<2 check is.

You should probably also raise a bug report with the ffmpeg crew about the lack of AudioSpecificConfig data in the avi header.


Also when putting files on dropbox, etc, for explicit download by others, rename them to have a .BIN file extension or wrap them in an archive format like .zip, .7z, .rar, etc. It will stop the smart arse dropbox site from wanting to play the files at you instead of offering a simple download.

[creaothceann]

Or just use mega.

[Wilbert]

You should probably also raise a bug report with the ffmpeg crew about the lack of AudioSpecificConfig data in the avi header.

Done: https://trac.ffmpeg.org/ticket/2883

@belonesox, you can also use AVIMux for muxing your avi and aac streams (it doesn't seem to write the bit depth though). It doesn't have this problem.