1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00

Original Commit: r92 | ods15 | 2006-09-30 12:13:47 +0300 (Sat, 30 Sep 2006) | 3 lines

user settable audio quality, with -aq. values between 0.1 to 0.3 are sane.
the higher the better.

Originally committed as revision 6496 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
Oded Shimon 2006-10-02 06:08:59 +00:00
parent b002eec985
commit 4000774d8f

View File

@ -114,6 +114,7 @@ typedef struct {
float * samples;
float * floor; // also used for tmp values for mdct
float * coeffs; // also used for residue after floor
float quality;
int ncodebooks;
codebook_t * codebooks;
@ -787,14 +788,14 @@ static void floor_fit(venc_context_t * venc, floor_t * fc, float * coeffs, int *
float tot_average = 0.;
for (i = 0; i < fc->values; i++) tot_average += get_floor_average(fc, coeffs, i);
tot_average /= fc->values;
tot_average /= 0.5;
tot_average /= venc->quality;
for (i = 0; i < fc->values; i++) {
int position = fc->list[fc->list[i].sort].x;
float average = get_floor_average(fc, coeffs, i);
int j;
average /= pow(average, 0.7) / tot_average * pow(0.9, position/200.); // MAGIC!
average /= pow(average, 0.5) / tot_average * pow(0.8, position/200.); // MAGIC!
for (j = 0; j < range - 1; j++) if (floor1_inverse_db_table[j * fc->multiplier] > average) break;
posts[fc->list[i].sort] = j;
}
@ -1064,7 +1065,8 @@ static int vorbis_encode_init(AVCodecContext * avccontext)
create_vorbis_context(venc, avccontext);
//if (avccontext->flags & CODEC_FLAG_QSCALE) avccontext->global_quality / (float)FF_QP2LAMBDA); else avccontext->bit_rate;
if (avccontext->flags & CODEC_FLAG_QSCALE) venc->quality = avccontext->global_quality / (float)FF_QP2LAMBDA;
else venc->quality = 0.17;
//if(avccontext->cutoff > 0) cfreq = avccontext->cutoff / 1000.0;
avccontext->extradata_size = put_main_header(venc, (uint8_t**)&avccontext->extradata);