[FFmpeg-devel] [PATCH] G.729A (now fixed-point) decoder

[Diego Biurrun]

On Sun, Mar 16, 2008 at 12:24:58AM +0600, Vladimir Voroshilov wrote:
> 
> Hopefully this version will be better.

build system part OK, some nits below

> --- /dev/null
> +++ b/libavcodec/g729dec.c
> @@ -0,0 +1,2010 @@
> +
> +   on presence of hearable artefacts/differences

artIfacts, same below, we are using American English here ;)

> +no prefix : common routines for miscelaneous tasks (e.g. fixed-point math operations)

miscellaneous, please keep lines below 80 chars, same below

> +-------------------------------------------------------------------------------
> +    Formats description

format

> +#define L0_BITS 1             ///< Switched MA predictor of LSP quantizer (size in bits)

Please lowercase all those lines.

> +#define L3_BITS 5             ///< First stage hihjer vector of quantizer (size in bits)

higher?

> +    uint8_t quantizer_2nd_hi; ///< First stage hihjer vector of quantizer (size in bits)

ditto

> + *   After some analisys i found this aproximation:

analysis, I

> + * Sslope used to compute y = cos(x)

Slope

> + * \brief multiplies 32-bit integer by abother 16-bit and divides result by 2^15

another

> + * \todo Better implementation requred

required

> + * \brief Decoding fo the fixed-codebook vector (3.8)

of?

> + * \return 1 if overflow occured, o - otherwise

0?

> + * \param residual (Q0) input data to filtering

filtering input data

> + * \param residual_filt [out] (Q0) speech signal with applied A(z/GAMMA_N) filter

Would it hurt to call this residual_filter?

> +    //Downscaling corellaions to fit on 16-bit

Downscale corellations to fit in 16 bit.

> +    /* 4.2.1, Equation 82. checking if filter should be disabled */

check

> +    /* 4.2.1, Equation 78, reconstructing delayed signal */

reconstruct

> +    /* A.4.2.3, Equation A.14, calcuating rh(0)  */

calculate

> +    /* A.4.2.3, Equation A.14, calcuating rh(1)  */

ditto

> +    /* A.4.2.3. Equation A.13, applying filter to signal */

apply

> +    // Copying data from previous frame

copy

> +    // Copying the rest of speech data

copy

> +    // Save data for using in next subframe

Save data to use it in the next subframe.

> + * Filtering has following  stages:

the following

> +    /* Calculating coefficients of A(z/GAMMA_N) filter */

Calculate A(z/GAMMA_N) filter coefficients

same/similar below

> + * \brief decodes polinomial coefficients from LSP

polynomial, same below


I'm getting bored at this point, but I'm sure there are a couple dozen
unnecessary -ing forms below :)

Diego