[FFmpeg-devel] [PATCH] E-AC-3 spectral extension
Sat Aug 15 19:24:07 CEST 2009
Justin Ruggles wrote:
> Michael Niedermayer wrote:
>> On Mon, Aug 03, 2009 at 06:08:52PM -0400, Justin Ruggles wrote:
>>> Michael Niedermayer wrote:
>>>>> + int8_t spx_atten_code[AC3_MAX_CHANNELS]; ///< spx attenuation code (spxattencod)
>>>>> + int spx_start_subband; ///< spx beginning frequency band (spxbegf)
>>>>> + int spx_start_freq; ///< spx start frequency bin
>>>>> + int spx_end_freq; ///< spx end frequency bin
>>>> what are frequency bins?
>>> MDCT coefficient index. Frequency bin is the term that the (E-)AC-3
>>> spec uses throughout.
>> Please explain these terms somewhere, like at the top of the file.
>> And dont hesitate to add diagrams and ascii art to make all the relations
> I have added a summary of the bin/subband/band grouping of MDCT
> coefficients used in (E-)AC-3...with an ascii art diagram. :)
>>>>> + int spx_copy_start_freq; ///< spx starting frequency for copying (copystartmant)
>>>> frequency in hz ?
>>> bin number. fixed.
>> next clarify what starting frequency is
>> there surely is a source and a destination to every copy ...
> the summary noted above defines what a frequency bin is. i also added a
> comment here stating that the copy range ends at the start of the spx range.
>>>>> + int num_spx_bands; ///< number of spx bands (nspxbnds)
>>>>> + uint8_t spx_band_struct[SPX_MAX_BANDS]; ///< spectral extension band structure (spxbndstrc)
>>>> what is a spectral extension band structure?
>>> determines how many subbands are in each band. for each subband, 1
>>> means combine with previous band, 0 means start new band. this is the
>>> same format as the coupling band structure, hence the shared function to
>>> decode it. I just added more information to the current documentation
>>> for decode_band_structure().
>> Could you please explain this in the doxy of the variable,
>> someone reading the code who sees "spx_band_struct" would look up
>> the doxy in the struct first and only if the doxy that is supposed
>> to explain does not do that at all would he read the code using the
>> you dont have to write all the details in the doxy but
>> "spectral extension band structure" says nothing unless one happens to
>> know the ac3 spec by heart and id love to not require that.
>> A person knowing how MDCT+VLC/RLE based audio coding works should be
>> able to make sense of such decoders in lavc and not have to read
>> through each individual spec
> I hope the recent changes I've made help this. I removed all the
> *_band_struct from the AC3DecodeContext. So what defines the band
> structure is the number of bands and size of each band, which is much
> more intuitive to someone not already knowing the spec.
> New patch attached.
> Thanks for the suggestions.
More information about the ffmpeg-devel