[FFmpeg-devel] Implementation of Huffman codes for DCA encoder

[Даниил Чередник]

Hi.
Is there any issue in this patch?

On Sun, Jan 8, 2017 at 1:22 AM, Даниил Чередник <dan.cherednik at gmail.com>
wrote:

> With real music and 256k bitrate encoding (the source was 44100, 16bit
> stereo) I got:
> Without Huffman: Best PSNR is  31.77 for shift 0
> With: Best PSNR is  37.45 for shift 0
>
> Current implementation of DCA encoder has minimal set of DTS features (no
> ADPCM, no VQ, fixed amount of transmitted subbands, no transient control,
> etc). So distortion is quite audible. Current set of patches introduces
> Huffman encoding for quantized audio data which is equivalent of increasing
> bitrate for 10-20%. Also bitstream allows to use Huffman for scale factor
> indexes, and some other data. I am working on it too.
>
> I have attached new split set of patches.
>
> On Sat, Jan 7, 2017 at 11:50 PM, Carl Eugen Hoyos <ceffmpeg at gmail.com>
> wrote:
>
>> 2017-01-07 20:39 GMT+01:00 Rostislav Pehlivanov <atomnuker at gmail.com>:
>> > On 7 January 2017 at 16:11, Carl Eugen Hoyos <ceffmpeg at gmail.com>
>> wrote:
>> >
>> >> 2017-01-07 16:00 GMT+01:00 Даниил Чередник <dan.cherednik at gmail.com>:
>> >>
>> >> > Currently I am working on improvement quality of DTS encoder.
>> Following
>> >> > patches introduce Huffman coding.
>> >>
>> >> Is the quality improvement so obvious that no further tests are
>> necessary?
>> >> (Does psnr improve measurably?)
>> >>
>> > PSNR is pretty much useless for audio.
>>
>> > The ear's the only metric which works.
>>
>> Not everybody's;-))
>>
>> > From the 2 samples he posted I can tell there's a big difference, and
>> the
>> > encoder isn't very good right now and the patch definitely helps.
>>
>> Thank you!
>>
>> Carl Eugen
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
>
>
>
> --
> Daniil Cherednik
>



-- 
Daniil Cherednik