[vlc-commits] audio_filter: add RNNoise denoiser filter
Tristan Matthews
git at videolan.org
Sat Oct 17 04:14:36 CEST 2020
vlc | branch: master | Tristan Matthews <tmatth at videolan.org> | Thu Jul 25 23:39:10 2019 -0400| [6416ea5448a9c16f3449b1c695acda5a356ce240] | committer: Tristan Matthews
audio_filter: add RNNoise denoiser filter
> http://git.videolan.org/gitweb.cgi/vlc.git/?a=commit;h=6416ea5448a9c16f3449b1c695acda5a356ce240
---
NEWS | 3 +
configure.ac | 5 ++
modules/audio_filter/Makefile.am | 8 ++
modules/audio_filter/rnnoise.c | 182 +++++++++++++++++++++++++++++++++++++++
po/POTFILES.in | 1 +
5 files changed, 199 insertions(+)
diff --git a/NEWS b/NEWS
index 50b8e7a9f6..f6e1a65c50 100644
--- a/NEWS
+++ b/NEWS
@@ -81,6 +81,9 @@ Video output:
* Remove Real demuxer plugin
* Fix washed out black on NVIDIA cards with Direct3D9
+Audio filter:
+ * Add RNNoise recurrent neural network denoiser
+
Video filter:
* Update yadif
* Remove remote OSD plugin
diff --git a/configure.ac b/configure.ac
index 55e4d17d86..4a19fdd999 100644
--- a/configure.ac
+++ b/configure.ac
@@ -3011,6 +3011,11 @@ AS_IF([test "${enable_speex}" != "no"], [
])
AM_CONDITIONAL([HAVE_SPEEXDSP], [test "$have_speexdsp" = "yes"])
+dnl
+dnl RNNoise plugin - Recurrent neural network for audio noise reduction
+dnl
+PKG_ENABLE_MODULES_VLC([RNNOISE], [], [rnnoise], [Rnnoise denoiser], [auto])
+
dnl
dnl Opus plugin
dnl
diff --git a/modules/audio_filter/Makefile.am b/modules/audio_filter/Makefile.am
index 028b188bfe..c77a90cb15 100644
--- a/modules/audio_filter/Makefile.am
+++ b/modules/audio_filter/Makefile.am
@@ -147,3 +147,11 @@ libspeex_resampler_plugin_la_LIBADD = $(SPEEXDSP_LIBS)
if HAVE_SPEEXDSP
audio_filter_LTLIBRARIES += libspeex_resampler_plugin.la
endif
+
+librnnoise_plugin_la_SOURCES = audio_filter/rnnoise.c
+librnnoise_plugin_la_CFLAGS = $(AM_CFLAGS) $(RNNOISE_CFLAGS)
+librnnoise_plugin_la_LIBADD = $(RNNOISE_LIBS) $(LIBM)
+librnnoise_plugin_la_LDFLAGS = $(AM_LDFLAGS) -rpath '$(audio_filterdir)'
+
+audio_filter_LTLIBRARIES += $(LTLIBrnnoise)
+EXTRA_LTLIBRARIES += librnnoise_plugin.la
diff --git a/modules/audio_filter/rnnoise.c b/modules/audio_filter/rnnoise.c
new file mode 100644
index 0000000000..34229f0012
--- /dev/null
+++ b/modules/audio_filter/rnnoise.c
@@ -0,0 +1,182 @@
+/*****************************************************************************
+ * rnnoise.c : Recurrent neural network for audio noise reduction
+ *****************************************************************************
+ * Copyright © 2019 Tristan Matthews
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU Lesser General Public License as published by
+ * the Free Software Foundation; either version 2.1 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
+ *****************************************************************************/
+
+#ifdef HAVE_CONFIG_H
+# include "config.h"
+#endif
+
+#include <vlc_common.h>
+#include <vlc_aout.h>
+#include <vlc_filter.h>
+#include <vlc_plugin.h>
+
+#include <rnnoise.h>
+
+typedef struct
+{
+ DenoiseState **p_sts;
+ float *p_scratch_buffer;
+ bool b_first;
+} filter_sys_t;
+
+static void
+Flush(filter_t *p_filter)
+{
+ filter_sys_t *p_sys = p_filter->p_sys;
+
+ if (p_sys->p_sts) {
+ int i_channels = p_filter->fmt_in.audio.i_channels;
+ for (int i = 0; i < i_channels; i++) {
+ rnnoise_destroy( p_sys->p_sts[i] );
+ }
+ free(p_sys->p_sts);
+ p_sys->p_sts = NULL;
+ p_sys->b_first = true;
+ }
+}
+
+static DenoiseState **init_denoise_state(unsigned i_channels)
+{
+ DenoiseState **p_sts = malloc(i_channels * sizeof(DenoiseState *));
+ if (unlikely(!p_sts))
+ return NULL;
+
+ for (unsigned i = 0; i < i_channels; i++)
+ {
+ p_sts[i] = rnnoise_create(NULL);
+ if (unlikely(!p_sts[i]))
+ {
+ for (unsigned j = 0; j < i; j++)
+ rnnoise_destroy( p_sts[j] );
+
+ return NULL;
+ }
+ }
+
+ return p_sts;
+}
+
+
+#define FRAME_SIZE 480
+
+static block_t *Process(filter_t *p_filter, block_t *p_block)
+{
+ filter_sys_t *p_sys = p_filter->p_sys;
+ float *p_buffer = (float *)p_block->p_buffer;
+ const int i_channels = p_filter->fmt_in.audio.i_channels;
+
+ float *tmp = p_sys->p_scratch_buffer;
+
+ if (unlikely(p_sys->p_sts == NULL))
+ {
+ /* Can happen after a flush */
+ p_sys->p_sts = init_denoise_state(i_channels);
+ if (p_sys->p_sts == NULL)
+ return p_block;
+ }
+
+ for (int i_nb_samples = p_block->i_nb_samples; i_nb_samples > 0; i_nb_samples -= FRAME_SIZE)
+ {
+ /* handle case where we have fewer than FRAME_SIZE samples left to process */
+ const unsigned frame_size = __MIN(FRAME_SIZE, i_nb_samples);
+
+ /* rnnoise processes blocks of 480 samples, and expects input to be in the 32768 scale. */
+ for (unsigned i = 0; i < frame_size; i++) {
+ for (int j = 0; j < i_channels; j++) {
+ tmp[i + j * frame_size] = p_buffer[i * i_channels + j] * 32768.f;
+ }
+ }
+
+ for (int i = 0; i < i_channels; i++) {
+ rnnoise_process_frame(p_sys->p_sts[i], tmp + i * frame_size , tmp + i * frame_size);
+ }
+
+ /* Skip writing first frame to output (as per the examples) I guess to prime rnnoise? */
+ if(!p_sys->b_first)
+ {
+ for (unsigned i = 0; i < frame_size; i++) {
+ for (int j = 0; j < i_channels; j++) {
+ p_buffer[i * i_channels + j] = tmp[i + j * frame_size] / 32768.f;
+ }
+ }
+ }
+ else
+ {
+ p_sys->b_first = false;
+ }
+ p_buffer += frame_size * i_channels;
+ }
+
+ return p_block;
+}
+
+static void Close( filter_t *obj )
+{
+ filter_t *p_filter = (filter_t *)obj;
+ filter_sys_t *p_sys = p_filter->p_sys;
+ Flush(p_filter);
+ free(p_sys->p_scratch_buffer);
+}
+
+static int Open (vlc_object_t *obj)
+{
+ filter_t *p_filter = (filter_t *)obj;
+
+ filter_sys_t *p_sys = p_filter->p_sys = vlc_obj_malloc(VLC_OBJECT(p_filter), sizeof(*p_sys));
+ if(unlikely(!p_sys))
+ return VLC_ENOMEM;
+
+ p_sys->b_first = true;
+
+ int i_channels = p_filter->fmt_in.audio.i_channels;
+ p_sys->p_sts = init_denoise_state(i_channels);
+ if (unlikely(!p_sys->p_sts)) {
+ vlc_obj_free(VLC_OBJECT(p_filter), p_sys);
+ return VLC_ENOMEM;
+ }
+
+ p_sys->p_scratch_buffer = malloc(FRAME_SIZE * i_channels * sizeof(*p_sys->p_scratch_buffer));
+ if (unlikely(!p_sys->p_scratch_buffer)) {
+ Flush(p_filter);
+ vlc_obj_free(VLC_OBJECT(p_filter), p_sys);
+ return VLC_ENOMEM;
+ }
+
+ p_filter->fmt_in.audio.i_format = VLC_CODEC_FL32;
+ aout_FormatPrepare(&p_filter->fmt_in.audio);
+ p_filter->fmt_out.audio = p_filter->fmt_in.audio;
+
+ static const struct vlc_filter_operations filter_ops =
+ {
+ .filter_audio = Process, .flush = Flush, .close = Close,
+ };
+ p_filter->ops = &filter_ops;
+
+ return VLC_SUCCESS;
+}
+
+vlc_module_begin ()
+ set_shortname (N_("RNNoise"))
+ set_description (N_("RNNoise filter"))
+ set_category (CAT_AUDIO)
+ set_subcategory (SUBCAT_AUDIO_AFILTER)
+ set_capability ("audio filter", 0)
+ set_callback( Open )
+vlc_module_end ()
diff --git a/po/POTFILES.in b/po/POTFILES.in
index 2432b295db..ddd566816a 100644
--- a/po/POTFILES.in
+++ b/po/POTFILES.in
@@ -247,6 +247,7 @@ modules/audio_filter/resampler/soxr.c
modules/audio_filter/resampler/speex.c
modules/audio_filter/resampler/src.c
modules/audio_filter/resampler/ugly.c
+modules/audio_filter/rnnoise.c
modules/audio_filter/scaletempo.c
modules/audio_filter/spatializer/spatializer.cpp
modules/audio_filter/stereo_widen.c
More information about the vlc-commits
mailing list