[pulseaudio-discuss] [PATCH v3 20/24] echo-cancel: Add beamforming support in the webrtc canceller
Arun Raghavan
arun at accosted.net
Wed Feb 17 13:17:17 UTC 2016
On Mon, 2016-02-15 at 13:12 +0200, Tanu Kaskinen wrote:
> On Mon, 2016-01-18 at 13:06 +0530, arun at accosted.net wrote:
> > From: Arun Raghavan <git at arunraghavan.net>
> >
> > ---
> > src/modules/echo-cancel/webrtc.cc | 138 +++++++++++++++++++++++++++++++++++++-
> > 1 file changed, 137 insertions(+), 1 deletion(-)
> >
> > diff --git a/src/modules/echo-cancel/webrtc.cc b/src/modules/echo-cancel/webrtc.cc
> > index 5f00286..9ae5d77 100644
> > --- a/src/modules/echo-cancel/webrtc.cc
> > +++ b/src/modules/echo-cancel/webrtc.cc
> > @@ -52,6 +52,7 @@ PA_C_DECL_END
> > #define DEFAULT_INTELLIGIBILITY_ENHANCER false
> > #define DEFAULT_EXPERIMENTAL_AGC false
> > #define DEFAULT_AGC_START_VOLUME 85
> > +#define DEFAULT_BEAMFORMING false
> > #define DEFAULT_TRACE false
> >
> > #define WEBRTC_AGC_MAX_VOLUME 255
> > @@ -70,6 +71,9 @@ static const char* const valid_modargs[] = {
> > "intelligibility_enhancer",
> > "experimental_agc",
> > "agc_start_volume",
> > + "beamforming",
> > + "mic_geometry", /* documented in parse_mic_geometry() */
> > + "target_direction", /* documented in parse_mic_geometry() */
> > "trace",
> > NULL
> > };
> > @@ -138,6 +142,76 @@ static void pa_webrtc_ec_fixate_spec(pa_sample_spec *rec_ss, pa_channel_map *rec
> > play_ss->rate = rec_ss->rate;
> > }
> >
> > +static bool parse_point(const char **point, float (&f)[3]) {
> > + int ret, length;
> > +
> > + ret = sscanf(*point, "%g,%g,%g%n", &f[0], &f[1], &f[2], &length);
> > + if (ret != 3)
> > + return false;
> > +
> > + /* Consume the bytes we've read so far */
> > + *point += length;
> > +
> > + /* Consume trailing comma if there is one */
> > + if (**point == ',')
> > + (*point)++;
>
> I don't think this should be done in this function. When expecting only
> one point, we should fail on "1,2,3," but this code accepts that
> string.
Okay.
> > +
> > + return true;
> > +}
> > +
> > +static bool parse_mic_geometry(const char **mic_geometry, std::vector& geometry) {
> > + /* The microphone geometry is expressed as cartesian point form:
> > + * x1,y1,z1,x2,y2,z2,...
> > + *
> > + * Where x1,y1,z1 is the position of the first microphone with regards to
> > + * the array's "center", x2,y2,z2 the position of the second, and so on.
> > + *
> > + * 'x' is the horizontal coordinate, with positive values being to the
> > + * right from the mic array's perspective.
> > + *
> > + * 'y' is the depth coordinate, with positive values being in front of the
> > + * array.
> > + *
> > + * 'z' is the vertical coordinate, with positive values being above the
> > + * array.
> > + *
> > + * All distances are in meters.
> > + */
>
> It's a bit unclear what "the mic array's perspective" means. Should I
> imagine the array as a person standing upright, looking straight ahead?
> Then positive x is to the right from the "person's" point of view,
> positive y is in front of the array guy and positive z is above the
> array.
Yes.
> > +
> > + /* The target direction is expected to be in spherical point form:
> > + * a,e,r
> > + *
> > + * Where 'a is the azimuth of the first mic channel, 'e' its elevation,
> > + * and 'r' the radius.
> > + *
> > + * 0 radians azimuth is to the right of the array, and positive angles
> > + * move in a counter-clockwise direction.
> > + *
> > + * 0 radians elevation is horizontal w.r.t. the array, and positive
> > + * angles go upwards.
> > + *
> > + * radius is distance from the array center in meters.
> > + */
> > +
> > + int i;
> > + float f[3];
> > +
> > + for (i = 0; i < geometry.size(); i++) {
> > + if (!parse_point(mic_geometry, f)) {
> > + pa_log("Failed to parse channel %d in mic_geometry", i);
> > + return false;
> > + }
> > +
> > + pa_log_debug("Got mic #%d position: (%g, %g, %g)", i, f[0], f[1], f[2]);
> > +
> > + geometry[i].c[0] = f[0];
> > + geometry[i].c[1] = f[1];
> > + geometry[i].c[2] = f[2];
> > + }
> > +
> > + return true;
> > +}
> > +
> > bool pa_webrtc_ec_init(pa_core *c, pa_echo_canceller *ec,
> > pa_sample_spec *rec_ss, pa_channel_map *rec_map,
> > pa_sample_spec *play_ss, pa_channel_map *play_map,
> > @@ -146,7 +220,7 @@ bool pa_webrtc_ec_init(pa_core *c, pa_echo_canceller *ec,
> > webrtc::AudioProcessing *apm = NULL;
> > webrtc::ProcessingConfig pconfig;
> > webrtc::Config config;
> > - bool hpf, ns, agc, dgc, mobile, cn, vad, ext_filter, intelligibility, experimental_agc;
> > + bool hpf, ns, agc, dgc, mobile, cn, vad, ext_filter, intelligibility, experimental_agc, beamforming;
> > int rm = -1;
> > uint32_t agc_start_volume;
> > pa_modargs *ma;
> > @@ -255,6 +329,12 @@ bool pa_webrtc_ec_init(pa_core *c, pa_echo_canceller *ec,
> > }
> > ec->params.webrtc.agc_start_volume = agc_start_volume;
> >
> > + beamforming = DEFAULT_BEAMFORMING;
> > + if (pa_modargs_get_value_boolean(ma, "beamforming", &beamforming) < 0) {
> > + pa_log("Failed to parse beamforming value");
> > + goto fail;
> > + }
> > +
> > if (ext_filter)
> > config.Set(new webrtc::ExtendedFilter(true));
> > if (intelligibility)
> > @@ -276,6 +356,62 @@ bool pa_webrtc_ec_init(pa_core *c, pa_echo_canceller *ec,
> >
> > pa_webrtc_ec_fixate_spec(rec_ss, rec_map, play_ss, play_map, out_ss, out_map);
> >
> > + /* We do this after fixate because we need the capture channel count */
> > + if (beamforming) {
> > + std::vector geometry(rec_ss->channels);
> > + webrtc::SphericalPointf direction(0.0f, 0.0f, 0.0f);
> > + const char *mic_geometry, *target_direction;
> > +
> > + if (!(mic_geometry = pa_modargs_get_value(ma, "mic_geometry", NULL))) {
> > + pa_log("mic_geometry must be set if beamforming is enabled");
> > + goto fail;
> > + }
> > +
> > + if (!parse_mic_geometry(&mic_geometry, geometry)) {
> > + pa_log("Failed to parse mic_geometry value");
> > + goto fail;
> > + }
> > +
> > + if (*mic_geometry != '\0') {
> > + pa_log("Failed to parse mic_geometry value: more parameters than expected");
> > + goto fail;
> > + }
>
> This check could be moved to parse_mic_geometry().
Fair enough.
> > +
> > + if ((target_direction = pa_modargs_get_value(ma, "target_direction", NULL))) {
> > + float f[3];
> > +
> > + if (!parse_point(&target_direction, f)) {
> > + pa_log("Failed to parse target_direction value");
> > + goto fail;
> > + }
> > +
> > + if (*target_direction != '\0') {
> > + pa_log("Failed to parse target_direction value: more parameters than expected");
> > + goto fail;
> > + }
> > +
> > +#define IS_ZERO(f) ((f) < 0.000001 && (f) > -0.000001)
>
> This kind of macro would be nice to have in some shared location.
> speex.c and a couple of tests have similar code, and I remember needing
> it in the volume control patches.
I can do that in a separate commit.
> > +
> > + if (!IS_ZERO(f[1]) || !IS_ZERO(f[2])) {
>
> This line contains a tab character.
Fixed.
> > + pa_log("The beamformer currently only supports targeting along the azimuth");
> > + goto fail;
> > + }
> > +
> > + direction.s[0] = f[0];
> > + direction.s[1] = f[1];
> > + direction.s[2] = f[2];
> > + }
> > +
> > + /* The beamformer gives us a single channel */
> > + out_ss->channels = 1;
> > + pa_channel_map_init_mono(out_map);
>
> Could this be moved to fixate_spec()? I don't like the audio format
> parameters being calculated all over the place.
I wanted to do that, but we can't because we need the capture channel
count for parsing the mic geometry (there is a comment explaining that
where if (beamforming) { ... } is).
-- Arun
More information about the pulseaudio-discuss
mailing list