generic-library/webrtc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

VAD refactoring: Added function for repeated code.

Browse Source 
Added WeightedAverage() to calculate global mean. This removes hard coded Gaussian model size and repeated code.

Tested with vad_unittests, audioproc_unittest and trybots.

BUG=None
TEST=None

Review URL: https://webrtc-codereview.appspot.com/571006

git-svn-id: http://webrtc.googlecode.com/svn/trunk@2275 4adac7df-926f-26a2-2b94-8c16560cd09d
This commit is contained in:
bjornv@webrtc.org 2012-05-23 07:56:51 +00:00
parent 2b004655b5
commit cb0a86e913
1 changed files with 36 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

72
src/common_audio/vad/vad_core.c
View File

@ -90,6 +90,26 @@ static const int16_t kOverHangMax2VAG[3] = { 9, 5, 3 };
static const int16_t kLocalThresholdVAG[3] = { 94, 94, 94 };
static const int16_t kGlobalThresholdVAG[3] = { 1100, 1050, 1100 };
// Calculates the weighted average w.r.t. number of Gaussians. The |data| are
// updated with an |offset| before averaging.
//
// - data [i/o] : Data to average.
// - offset [i] : An offset added to |data|.
// - weights [i] : Weights used for averaging.
//
// returns : The weighted average.
static int32_t WeightedAverage(int16_t* data, int16_t offset,
const int16_t* weights) {
int k;
int32_t weighted_average = 0;
for (k = 0; k < kNumGaussians; k++) {
data[k * kNumChannels] += offset;
weighted_average += data[k * kNumChannels] * weights[k * kNumChannels];
}
return weighted_average;
}
// Calculates the probabilities for both speech and background noise using
// Gaussian Mixture Models (GMM). A hypothesis-test is performed to decide which
// type of signal is most probable.
@ -241,11 +261,8 @@ static int16_t GmmProbability(VadInstT* self, int16_t* feature_vector,
feature_minimum = WebRtcVad_FindMinimum(self, feature_vector[n], n);
// Compute the "global" mean, that is the sum of the two means weighted.
noise_global_mean = WEBRTC_SPL_MUL_16_16(kNoiseDataWeights[n],
*nmean1ptr); // Q7 * Q7
noise_global_mean +=
WEBRTC_SPL_MUL_16_16(kNoiseDataWeights[n + kNumChannels],
*(nmean1ptr + kNumChannels));
noise_global_mean = WeightedAverage(&self->noise_means[n], 0,
&kNoiseDataWeights[n]);
tmp1_s16 = (int16_t) (noise_global_mean >> 6); // Q8
for (k = 0; k < kNumGaussians; k++) {
@ -381,18 +398,12 @@ static int16_t GmmProbability(VadInstT* self, int16_t* feature_vector,
// Separate models if they are too close.
// |noise_global_mean| in Q14 (= Q7 * Q7).
noise_global_mean = WEBRTC_SPL_MUL_16_16(kNoiseDataWeights[n],
*nmean1ptr);
noise_global_mean +=
WEBRTC_SPL_MUL_16_16(kNoiseDataWeights[n + kNumChannels],
*nmean2ptr);
noise_global_mean = WeightedAverage(&self->noise_means[n], 0,
&kNoiseDataWeights[n]);
// |speech_global_mean| in Q14 (= Q7 * Q7).
speech_global_mean = WEBRTC_SPL_MUL_16_16(kSpeechDataWeights[n],
*smean1ptr);
speech_global_mean +=
WEBRTC_SPL_MUL_16_16(kSpeechDataWeights[n + kNumChannels],
*smean2ptr);
speech_global_mean = WeightedAverage(&self->speech_means[n], 0,
&kSpeechDataWeights[n]);
// |diff| = "global" speech mean - "global" noise mean.
// (Q14 >> 9) - (Q14 >> 9) = Q5.
@ -406,28 +417,17 @@ static int16_t GmmProbability(VadInstT* self, int16_t* feature_vector,
tmp1_s16 = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(13, tmp_s16, 2);
tmp2_s16 = (int16_t) WEBRTC_SPL_MUL_16_16_RSFT(3, tmp_s16, 2);
// First Gaussian, speech model.
tmp_s16 = tmp1_s16 + *smean1ptr;
*smean1ptr = tmp_s16;
speech_global_mean = WEBRTC_SPL_MUL_16_16(tmp_s16,
kSpeechDataWeights[n]);
// Move Gaussian means for speech model by |tmp1_s16| and update
// |speech_global_mean|. Note that |self->speech_means[n]| is changed
// after the call.
speech_global_mean = WeightedAverage(&self->speech_means[n], tmp1_s16,
&kSpeechDataWeights[n]);
// Second Gaussian, speech model.
tmp_s16 = tmp1_s16 + *smean2ptr;
*smean2ptr = tmp_s16;
speech_global_mean +=
WEBRTC_SPL_MUL_16_16(tmp_s16, kSpeechDataWeights[n + kNumChannels]);
// First Gaussian, noise model.
tmp_s16 = *nmean1ptr - tmp2_s16;
*nmean1ptr = tmp_s16;
noise_global_mean = WEBRTC_SPL_MUL_16_16(tmp_s16, kNoiseDataWeights[n]);
// Second Gaussian, noise model.
tmp_s16 = *nmean2ptr - tmp2_s16;
*nmean2ptr = tmp_s16;
noise_global_mean +=
WEBRTC_SPL_MUL_16_16(tmp_s16, kNoiseDataWeights[n + kNumChannels]);
// Move Gaussian means for noise model by -|tmp2_s16| and update
// |noise_global_mean|. Note that |self->noise_means[n]| is changed
// after the call.
noise_global_mean = WeightedAverage(&self->noise_means[n], -tmp2_s16,
&kNoiseDataWeights[n]);
}
// Control that the speech & noise means do not drift to much.