| Message ID | D44228D8-1F69-49A8-8E87-E055F5F29161@raphael.schlarb.one |
|---|---|
| State | New |
| Headers | show |
| Series | [FFmpeg-devel] Add option to encode short aac ld/eld frames with libfdk_aac | expand |
| Context | Check | Description |
|---|---|---|
| andriy/configure_x86 | warning | Failed to apply patch |
On 3/27/2023 9:28 AM, Raphael Schlarb wrote: > Some specifications require the size of ld/eld frames to be 480 samples > instead of the default 512. libfdk_aac provides an option to set an alternative > frame size, but it's not exposed via the ffmpeg interface. > This patch adds a boolean short_frame option to encode ld/eld frames of > size 480. > > Signed-off-by: Raphael Schlarb <info@raphael.schlarb.one> > --- > libavcodec/libfdk-aacenc.c | 13 +++++++++++++ > 1 file changed, 13 insertions(+) > > diff --git a/libavcodec/libfdk-aacenc.c b/libavcodec/libfdk-aacenc.c > index eb97e0fb41..bd719850dd 100644 > --- a/libavcodec/libfdk-aacenc.c > +++ b/libavcodec/libfdk-aacenc.c > @@ -55,6 +55,7 @@ typedef struct AACContext { > int metadata_mode; > AACENC_MetaData metaDataSetup; > int delay_sent; > + int short_frame; > > AudioFrameQueue afq; > } AACContext; > @@ -78,6 +79,7 @@ static const AVOption aac_enc_options[] = { > { "comp_profile", "The desired compression profile for AAC DRC", offsetof(AACContext, comp_profile), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 256, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM }, > { "comp_target_ref", "Expected target reference level at decoder side in dB (for clipping prevention/limiter)", offsetof(AACContext, comp_target_ref), AV_OPT_TYPE_INT, { .i64 = 0.0 }, -31.75, 0, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM }, > { "prog_ref", "The program reference level or dialog level in dB", offsetof(AACContext, prog_ref), AV_OPT_TYPE_INT, { .i64 = 0.0 }, -31.75, 0, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM }, > + { "short_frame", "Encode short LD/ELD frames, using 480 instead of 512 samples per frame", offsetof(AACContext, short_frame), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM }, > FF_AAC_PROFILE_OPTS > { NULL } > }; > @@ -166,6 +168,17 @@ static av_cold int aac_encode_init(AVCodecContext *avctx) > } > } > > + if (s->short_frame) { > + if (aot == FF_PROFILE_AAC_LD + 1 || aot == FF_PROFILE_AAC_ELD + 1) { > + if ((err = aacEncoder_SetParam(s->handle, AACENC_GRANULE_LENGTH, > + 480)) != AACENC_OK) { Why hardcode it to 480 if the parameter accepts arbitrary values? The documentation states the currently supported values are 120, 128, 240, 256, 480, 512 and 1024. Call the option frame_length, and set the range to 120-1024, with 1024 as default value. Then in here change it to 512 if it's 1024 and profile is FF_PROFILE_AAC_LD or FF_PROFILE_AAC_ELD (otherwise don't change it), always set it, and have the library reject or accept the value. > + av_log(avctx, AV_LOG_ERROR, "Unable to set granule length: %s\n", > + aac_get_error(err)); > + goto error; > + } > + } > + } > + > if ((err = aacEncoder_SetParam(s->handle, AACENC_SAMPLERATE, > avctx->sample_rate)) != AACENC_OK) { > av_log(avctx, AV_LOG_ERROR, "Unable to set the sample rate %d: %s\n",
diff --git a/libavcodec/libfdk-aacenc.c b/libavcodec/libfdk-aacenc.c index eb97e0fb41..bd719850dd 100644 --- a/libavcodec/libfdk-aacenc.c +++ b/libavcodec/libfdk-aacenc.c @@ -55,6 +55,7 @@ typedef struct AACContext { int metadata_mode; AACENC_MetaData metaDataSetup; int delay_sent; + int short_frame; AudioFrameQueue afq; } AACContext; @@ -78,6 +79,7 @@ static const AVOption aac_enc_options[] = { { "comp_profile", "The desired compression profile for AAC DRC", offsetof(AACContext, comp_profile), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 256, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM }, { "comp_target_ref", "Expected target reference level at decoder side in dB (for clipping prevention/limiter)", offsetof(AACContext, comp_target_ref), AV_OPT_TYPE_INT, { .i64 = 0.0 }, -31.75, 0, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM }, { "prog_ref", "The program reference level or dialog level in dB", offsetof(AACContext, prog_ref), AV_OPT_TYPE_INT, { .i64 = 0.0 }, -31.75, 0, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM }, + { "short_frame", "Encode short LD/ELD frames, using 480 instead of 512 samples per frame", offsetof(AACContext, short_frame), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM }, FF_AAC_PROFILE_OPTS { NULL } }; @@ -166,6 +168,17 @@ static av_cold int aac_encode_init(AVCodecContext *avctx) } } + if (s->short_frame) { + if (aot == FF_PROFILE_AAC_LD + 1 || aot == FF_PROFILE_AAC_ELD + 1) { + if ((err = aacEncoder_SetParam(s->handle, AACENC_GRANULE_LENGTH, + 480)) != AACENC_OK) { + av_log(avctx, AV_LOG_ERROR, "Unable to set granule length: %s\n", + aac_get_error(err)); + goto error; + } + } + } + if ((err = aacEncoder_SetParam(s->handle, AACENC_SAMPLERATE, avctx->sample_rate)) != AACENC_OK) { av_log(avctx, AV_LOG_ERROR, "Unable to set the sample rate %d: %s\n",
Some specifications require the size of ld/eld frames to be 480 samples instead of the default 512. libfdk_aac provides an option to set an alternative frame size, but it's not exposed via the ffmpeg interface. This patch adds a boolean short_frame option to encode ld/eld frames of size 480. Signed-off-by: Raphael Schlarb <info@raphael.schlarb.one> --- libavcodec/libfdk-aacenc.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)