[Libav-user] Audio quality loss while encoding

[Paul B Mahol]

On 4/24/13, Brad O'Hearne <brado at bighillsoftware.com> wrote:
> On Apr 24, 2013, at 8:30 AM, Claudio Freire <klaussfreire at gmail.com> wrote:
>
>> On Wed, Apr 24, 2013 at 8:21 AM, Pradeep Karosiya <praks411 at gmail.com>
>> wrote:
>>> The codec id used is AV_CODEC_ID_AAC. The encoding goes fine but I'm
>>> getting
>>> some noise is the final audio output. The quality is getting degraded.
>>> However if the number of channels is 1 then I'm getting same quality as
>>> the
>>> input file
>>
>>
>> I also experienced this when encoding AAC, so I stopped using AAC.
>>
>> I'm not using the very latest ffmpeg, so I didn't bother reporting the
>> issue, and it's been on my To-Do to check the latest version for a
>> while. Been busy and all. Anyway, I thought I'd say "me too", and add
>> that it does seem related to the AAC encoder, because it only happens
>> with AAC.
>
> This seems like some level of vindication / relief for this mortal -- as is
> probably apparent from my other thread, this is the exact problem I am
> having too. I am heavily suspecting weirdness in either resampling, or the
> nature of the expected data buffer to the encoder -- that's actually
> contained within a frame, so the data filling occurs within this function
> call:
>
> avcodec_fill_audio_frame(_streamAudioFrame,
>
> _streamAudioFrame->channels,
>
> _streamAudioFrame->format,
>                                                      (const uint8_t
> *)destinationData[0],
>                                                      bufferSize,
>                                                      0);
>
> To revisit this, my destinationData variable is declared as:
>
>                 uint8_t **destinationData = NULL;
>
> It is populated with data from resampling as in:
>
>                 returnVal = swr_convert(resamplerCtx,
>                                         destinationData,
>                                         destinationNumberOfSamples,
>                                         (const uint8_t **)sourceData,
>                                         sourceNumberOfSamples);
>
> from a source data array declared similarly as:
>
>                 uint8_t **sourceData = NULL;
>
> My questions revolve around this parameter in the avcodec_fill_audio_frame
> function:
>
>                  (const uint8_t *)destinationData[0],
>
> 1. Is that the proper pointer to be passing? It works (to the degree that
> the return code is success). I've tried sending:
>
> 		(const uint8_t *)destinationData,
>
> and
>
> 		(const uint8_t *)&destinationData,
>
> and
>
>                  (const uint8_t *)&destinationData[0],
>
> and all three of those blow up. I wouldn't otherwise doubt the original line
> (as it produced good audio with another sample format), but I've got little
> in the equation here that can be changed, so I'm reconsidering everything.
>
> 2. For planar data, what exactly is the expected structure of the data
> array?
>
> is it:
>
> data[0] -> plane 1
> data[1] -> plane 2

For planar sample format, each channel is in separate array:

channel left = data[0]
channel right = data[1]

If you use more than 8 channels, use extended_data instead.