Adds possibility to log delay estimates in AEC.

Review URL: http://webrtc-codereview.appspot.com/178001

git-svn-id: http://webrtc.googlecode.com/svn/trunk@674 4adac7df-926f-26a2-2b94-8c16560cd09d

src/modules/audio_processing/aec/main/interface/echo_cancellation.h

@@ -38,6 +38,7 @@ typedef struct {
     WebRtc_Word16 nlpMode;        // default kAecNlpModerate
     WebRtc_Word16 skewMode;       // default kAecFalse
     WebRtc_Word16 metricsMode;    // default kAecFalse
+    int delay_logging;            // default kAecFalse
     //float realSkew;
 } AecConfig;
 
@@ -66,7 +67,7 @@ extern "C" {
  * Inputs                       Description
  * -------------------------------------------------------------------
  * void **aecInst               Pointer to the AEC instance to be created
- *                              and initilized
+ *                              and initialized
  *
  * Outputs                      Description
  * -------------------------------------------------------------------
@@ -225,6 +226,23 @@ WebRtc_Word32 WebRtcAec_get_echo_status(void *aecInst, WebRtc_Word16 *status);
  */
 WebRtc_Word32 WebRtcAec_GetMetrics(void *aecInst, AecMetrics *metrics);
 
+/*
+ * Gets the current delay metrics for the session.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void*      handle            Pointer to the AEC instance
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * int*       median            Delay median value.
+ * int*       std               Delay standard deviation.
+ *
+ * int        return             0: OK
+ *                              -1: error
+ */
+int WebRtcAec_GetDelayMetrics(void* handle, int* median, int* std);
+
 /*
  * Gets the last error code.
  *

src/modules/audio_processing/aec/main/source/aec_core.c

@@ -12,12 +12,14 @@
  * The core AEC algorithm, which is presented with time-aligned signals.
  */
 
+#include "aec_core.h"
+
 #include <math.h>
 #include <stdlib.h>
 #include <string.h>
 
-#include "aec_core.h"
 #include "aec_rdft.h"
+#include "delay_estimator_float.h"
 #include "ring_buffer.h"
 #include "system_wrappers/interface/cpu_features_wrapper.h"
 
@@ -174,6 +176,15 @@ int WebRtcAec_CreateAec(aec_t **aecInst)
         return -1;
     }
 
+    if (WebRtc_CreateDelayEstimatorFloat(&aec->delay_estimator,
+                                         PART_LEN1,
+                                         kMaxDelay,
+                                         0) == -1) {
+      WebRtcAec_FreeAec(aec);
+      aec = NULL;
+      return -1;
+    }
+
     return 0;
 }
 
@@ -190,6 +201,8 @@ int WebRtcAec_FreeAec(aec_t *aec)
     WebRtcApm_FreeBuffer(aec->nearFrBufH);
     WebRtcApm_FreeBuffer(aec->outFrBufH);
 
+    WebRtc_FreeDelayEstimatorFloat(aec->delay_estimator);
+
     free(aec);
     return 0;
 }
@@ -371,6 +384,12 @@ int WebRtcAec_InitAec(aec_t *aec, int sampFreq)
         return -1;
     }
 
+    if (WebRtc_InitDelayEstimatorFloat(aec->delay_estimator) != 0) {
+      return -1;
+    }
+    aec->delay_logging_enabled = 0;
+    memset(aec->delay_histogram, 0, sizeof(aec->delay_histogram));
+
     // Default target suppression level
     aec->targetSupp = -11.5;
     aec->minOverDrive = 2.0;
@@ -561,6 +580,10 @@ static void ProcessBlock(aec_t *aec, const short *farend,
     float fft[PART_LEN2];
     float xf[2][PART_LEN1], yf[2][PART_LEN1], ef[2][PART_LEN1];
     complex_t df[PART_LEN1];
+    float far_spectrum = 0.0f;
+    float near_spectrum = 0.0f;
+    float abs_far_spectrum[PART_LEN1];
+    float abs_near_spectrum[PART_LEN1];
 
     const float gPow[2] = {0.9f, 0.1f};
 
@@ -625,10 +648,15 @@ static void ProcessBlock(aec_t *aec, const short *farend,
 
     // Power smoothing
     for (i = 0; i < PART_LEN1; i++) {
-        aec->xPow[i] = gPow[0] * aec->xPow[i] + gPow[1] * NR_PART *
-            (xf[0][i] * xf[0][i] + xf[1][i] * xf[1][i]);
-        aec->dPow[i] = gPow[0] * aec->dPow[i] + gPow[1] *
-            (df[i][0] * df[i][0] + df[i][1] * df[i][1]);
+      far_spectrum = xf[0][i] * xf[0][i] + xf[1][i] * xf[1][i];
+      aec->xPow[i] = gPow[0] * aec->xPow[i] + gPow[1] * NR_PART * far_spectrum;
+      // Calculate absolute spectra
+      abs_far_spectrum[i] = sqrtf(far_spectrum);
+
+      near_spectrum = df[i][0] * df[i][0] + df[i][1] * df[i][1];
+      aec->dPow[i] = gPow[0] * aec->dPow[i] + gPow[1] * near_spectrum;
+      // Calculate absolute spectra
+      abs_near_spectrum[i] = sqrtf(near_spectrum);
     }
 
     // Estimate noise power. Wait until dPow is more stable.
@@ -663,6 +691,20 @@ static void ProcessBlock(aec_t *aec, const short *farend,
         aec->noisePow = aec->dMinPow;
     }
 
+    // Block wise delay estimation used for logging
+    if (aec->delay_logging_enabled) {
+      int delay_estimate = 0;
+      // Estimate the delay
+      delay_estimate = WebRtc_DelayEstimatorProcessFloat(aec->delay_estimator,
+                                                         abs_far_spectrum,
+                                                         abs_near_spectrum,
+                                                         PART_LEN1,
+                                                         aec->echoState);
+      if (delay_estimate >= 0) {
+        // Update delay estimate buffer
+        aec->delay_histogram[delay_estimate]++;
+      }
+    }
 
     // Update the xfBuf block position.
     aec->xfBufBlockPos--;

src/modules/audio_processing/aec/main/source/aec_core.h

@@ -33,6 +33,8 @@
 #define PREF_BAND_SIZE 24
 
 #define BLOCKL_MAX FRAME_LEN
+// Maximum delay in fixed point delay estimator, used for logging
+enum {kMaxDelay = 100};
 
 typedef float complex_t[2];
 // For performance reasons, some arrays of complex numbers are replaced by twice
@@ -141,6 +143,10 @@ typedef struct {
     int flag_Hband_cn;      //for comfort noise
     float cn_scale_Hband;   //scale for comfort noise in H band
 
+    int delay_histogram[kMaxDelay];
+    int delay_logging_enabled;
+    void* delay_estimator;
+
 #ifdef AEC_DEBUG
     FILE *farFile;
     FILE *nearFile;

src/modules/audio_processing/aec/main/source/echo_cancellation.c

@@ -11,16 +11,18 @@
 /*
  * Contains the API functions for the AEC.
  */
+#include "echo_cancellation.h"
+
+#include <math.h>
+#ifdef AEC_DEBUG
+#include <stdio.h>
+#endif
 #include <stdlib.h>
 #include <string.h>
 
-#include "echo_cancellation.h"
 #include "aec_core.h"
-#include "ring_buffer.h"
 #include "resampler.h"
-#ifdef AEC_DEBUG
-    #include <stdio.h>
-#endif
+#include "ring_buffer.h"
 
 #define BUF_SIZE_FRAMES 50 // buffer size (frames)
 // Maximum length of resampled signal. Must be an integer multiple of frames
@@ -215,7 +217,7 @@ WebRtc_Word32 WebRtcAec_Init(void *aecInst, WebRtc_Word32 sampFreq, WebRtc_Word3
         return -1;
     }
 
-    aecpc->initFlag = initCheck;  // indicates that initilisation has been done
+    aecpc->initFlag = initCheck;  // indicates that initialization has been done
 
     if (aecpc->sampFreq == 32000) {
         aecpc->splitSampFreq = 16000;
@@ -254,6 +256,7 @@ WebRtc_Word32 WebRtcAec_Init(void *aecInst, WebRtc_Word32 sampFreq, WebRtc_Word3
     aecConfig.nlpMode = kAecNlpModerate;
     aecConfig.skewMode = kAecFalse;
     aecConfig.metricsMode = kAecFalse;
+    aecConfig.delay_logging = kAecFalse;
 
     if (WebRtcAec_set_config(aecpc, aecConfig) == -1) {
         aecpc->lastError = AEC_UNSPECIFIED_ERROR;
@@ -566,6 +569,15 @@ WebRtc_Word32 WebRtcAec_set_config(void *aecInst, AecConfig config)
         WebRtcAec_InitMetrics(aecpc->aec);
     }
 
+  if (config.delay_logging != kAecFalse && config.delay_logging != kAecTrue) {
+    aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
+    return -1;
+  }
+  aecpc->aec->delay_logging_enabled = config.delay_logging;
+  if (aecpc->aec->delay_logging_enabled == kAecTrue) {
+    memset(aecpc->aec->delay_histogram, 0, sizeof(aecpc->aec->delay_histogram));
+  }
+
     return 0;
 }
 
@@ -590,6 +602,7 @@ WebRtc_Word32 WebRtcAec_get_config(void *aecInst, AecConfig *config)
     config->nlpMode = aecpc->nlpMode;
     config->skewMode = aecpc->skewMode;
     config->metricsMode = aecpc->aec->metricsMode;
+    config->delay_logging = aecpc->aec->delay_logging_enabled;
 
     return 0;
 }
@@ -717,6 +730,69 @@ WebRtc_Word32 WebRtcAec_GetMetrics(void *aecInst, AecMetrics *metrics)
     return 0;
 }
 
+int WebRtcAec_GetDelayMetrics(void* handle, int* median, int* std) {
+  aecpc_t* self = handle;
+  int i = 0;
+  int delay_values = 0;
+  int num_delay_values = 0;
+  int my_median = 0;
+  float l1_norm = 0;
+
+  if (self == NULL) {
+    return -1;
+  }
+  if (median == NULL) {
+    self->lastError = AEC_NULL_POINTER_ERROR;
+    return -1;
+  }
+  if (std == NULL) {
+    self->lastError = AEC_NULL_POINTER_ERROR;
+    return -1;
+  }
+  if (self->initFlag != initCheck) {
+    self->lastError = AEC_UNINITIALIZED_ERROR;
+    return -1;
+  }
+  if (self->aec->delay_logging_enabled == 0) {
+    // Logging disabled
+    self->lastError = AEC_UNSUPPORTED_FUNCTION_ERROR;
+    return -1;
+  }
+
+  // Get number of delay values since last update
+  for (i = 0; i < kMaxDelay; i++) {
+    num_delay_values += self->aec->delay_histogram[i];
+  }
+  if (num_delay_values == 0) {
+    // We have no new delay value data
+    *median = -1;
+    *std = -1;
+    return 0;
+  }
+
+  delay_values = num_delay_values >> 1; // Start value for median count down
+  // Get median of delay values since last update
+  for (i = 0; i < kMaxDelay; i++) {
+    delay_values -= self->aec->delay_histogram[i];
+    if (delay_values < 0) {
+      my_median = i;
+      break;
+    }
+  }
+  *median = my_median;
+
+  // Calculate the L1 norm, with median value as central moment
+  for (i = 0; i < kMaxDelay; i++) {
+    l1_norm += (float) (fabs(i - my_median) * self->aec->delay_histogram[i]);
+  }
+  *std = (int) (l1_norm / (float) num_delay_values + 0.5f);
+
+  // Reset histogram
+  memset(self->aec->delay_histogram, 0, sizeof(self->aec->delay_histogram));
+
+  return 0;
+}
+
 WebRtc_Word32 WebRtcAec_get_version(WebRtc_Word8 *versionStr, WebRtc_Word16 len)
 {
     const char version[] = "AEC 2.5.0";

src/modules/audio_processing/main/interface/audio_processing.h

@@ -322,6 +322,16 @@ class EchoCancellation {
   // TODO(ajm): discuss the metrics update period.
   virtual int GetMetrics(Metrics* metrics) = 0;
 
+  // Enables computation and logging of delay values. Statistics are obtained
+  // through |GetDelayMetrics()|.
+  virtual int enable_delay_logging(bool enable) = 0;
+  virtual bool is_delay_logging_enabled() const = 0;
+
+  // The delay metrics consists of the delay |median| and the delay standard
+  // deviation |std|. The values are averaged over the time period since the
+  // last call to |GetDelayMetrics()|.
+  virtual int GetDelayMetrics(int* median, int* std) = 0;
+
  protected:
   virtual ~EchoCancellation() {};
 };

src/modules/audio_processing/main/source/echo_cancellation_impl.cc

@@ -66,7 +66,8 @@ EchoCancellationImpl::EchoCancellationImpl(const AudioProcessingImpl* apm)
     device_sample_rate_hz_(48000),
     stream_drift_samples_(0),
     was_stream_drift_set_(false),
-    stream_has_echo_(false) {}
+    stream_has_echo_(false),
+    delay_logging_enabled_(false) {}
 
 EchoCancellationImpl::~EchoCancellationImpl() {}
 
@@ -283,6 +284,39 @@ bool EchoCancellationImpl::stream_has_echo() const {
   return stream_has_echo_;
 }
 
+int EchoCancellationImpl::enable_delay_logging(bool enable) {
+  CriticalSectionScoped crit_scoped(*apm_->crit());
+  delay_logging_enabled_ = enable;
+  return Configure();
+}
+
+bool EchoCancellationImpl::is_delay_logging_enabled() const {
+  return delay_logging_enabled_;
+}
+
+// TODO(bjornv): How should we handle the multi-channel case?
+int EchoCancellationImpl::GetDelayMetrics(int* median, int* std) {
+  CriticalSectionScoped crit_scoped(*apm_->crit());
+  if (median == NULL) {
+    return apm_->kNullPointerError;
+  }
+  if (std == NULL) {
+    return apm_->kNullPointerError;
+  }
+
+  if (!is_component_enabled() || !delay_logging_enabled_) {
+    return apm_->kNotEnabledError;
+  }
+
+  Handle* my_handle = static_cast<Handle*>(handle(0));
+  if (WebRtcAec_GetDelayMetrics(my_handle, median, std) !=
+      apm_->kNoError) {
+    return GetHandleError(my_handle);
+  }
+
+  return apm_->kNoError;
+}
+
 int EchoCancellationImpl::Initialize() {
   int err = ProcessingComponent::Initialize();
   if (err != apm_->kNoError || !is_component_enabled()) {
@@ -332,6 +366,7 @@ int EchoCancellationImpl::ConfigureHandle(void* handle) const {
   config.metricsMode = metrics_enabled_;
   config.nlpMode = MapSetting(suppression_level_);
   config.skewMode = drift_compensation_enabled_;
+  config.delay_logging = delay_logging_enabled_;
 
   return WebRtcAec_set_config(static_cast<Handle*>(handle), config);
 }

src/modules/audio_processing/main/source/echo_cancellation_impl.h

@@ -49,6 +49,9 @@ class EchoCancellationImpl : public EchoCancellation,
   virtual bool are_metrics_enabled() const;
   virtual bool stream_has_echo() const;
   virtual int GetMetrics(Metrics* metrics);
+  virtual int enable_delay_logging(bool enable);
+  virtual bool is_delay_logging_enabled() const;
+  virtual int GetDelayMetrics(int* median, int* std);
 
   // ProcessingComponent implementation.
   virtual void* CreateHandle() const;
@@ -66,6 +69,7 @@ class EchoCancellationImpl : public EchoCancellation,
   int stream_drift_samples_;
   bool was_stream_drift_set_;
   bool stream_has_echo_;
+  bool delay_logging_enabled_;
 };
 }  // namespace webrtc
 

src/modules/audio_processing/main/test/process_test/process_test.cc

@@ -93,8 +93,8 @@ void usage() {
   printf("\n  -aec     Echo cancellation\n");
   printf("  --drift_compensation\n");
   printf("  --no_drift_compensation\n");
-  printf("  --echo_metrics\n");
   printf("  --no_echo_metrics\n");
+  printf("  --no_delay_logging\n");
   printf("\n  -aecm    Echo control mobile\n");
   printf("  --aecm_echo_path_in_file FILE\n");
   printf("  --aecm_echo_path_out_file FILE\n");
@@ -218,6 +218,10 @@ void void_main(int argc, char* argv[]) {
 
     } else if (strcmp(argv[i], "-aec") == 0) {
       ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true));
+      ASSERT_EQ(apm->kNoError,
+                apm->echo_cancellation()->enable_metrics(true));
+      ASSERT_EQ(apm->kNoError,
+                apm->echo_cancellation()->enable_delay_logging(true));
 
     } else if (strcmp(argv[i], "--drift_compensation") == 0) {
       ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true));
@@ -230,16 +234,16 @@ void void_main(int argc, char* argv[]) {
       ASSERT_EQ(apm->kNoError,
                 apm->echo_cancellation()->enable_drift_compensation(false));
 
-    } else if (strcmp(argv[i], "--echo_metrics") == 0) {
-      ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true));
-      ASSERT_EQ(apm->kNoError,
-                apm->echo_cancellation()->enable_metrics(true));
-
     } else if (strcmp(argv[i], "--no_echo_metrics") == 0) {
       ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true));
       ASSERT_EQ(apm->kNoError,
                 apm->echo_cancellation()->enable_metrics(false));
 
+    } else if (strcmp(argv[i], "--no_delay_logging") == 0) {
+      ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true));
+      ASSERT_EQ(apm->kNoError,
+                apm->echo_cancellation()->enable_delay_logging(false));
+
     } else if (strcmp(argv[i], "-aecm") == 0) {
       ASSERT_EQ(apm->kNoError, apm->echo_control_mobile()->Enable(true));
 
@@ -884,6 +888,14 @@ void void_main(int argc, char* argv[]) {
       printf("ANLP: ");
       PrintStat(metrics.a_nlp);
     }
+    if (apm->echo_cancellation()->is_delay_logging_enabled()) {
+      int median = 0;
+      int std = 0;
+      apm->echo_cancellation()->GetDelayMetrics(&median, &std);
+      printf("\n--Delay metrics--\n");
+      printf("Median:             %3d\n", median);
+      printf("Standard deviation: %3d\n", std);
+    }
   }
 
   if (!pb_file) {

src/modules/audio_processing/main/test/unit_test/unit_test.cc

@@ -457,6 +457,8 @@ TEST_F(ApmTest, Process) {
             apm_->echo_cancellation()->enable_drift_compensation(true));
   EXPECT_EQ(apm_->kNoError,
             apm_->echo_cancellation()->enable_metrics(true));
+  EXPECT_EQ(apm_->kNoError,
+            apm_->echo_cancellation()->enable_delay_logging(true));
   EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(true));
 
   EXPECT_EQ(apm_->kNoError,
@@ -591,6 +593,10 @@ TEST_F(ApmTest, Process) {
     EchoCancellation::Metrics echo_metrics;
     EXPECT_EQ(apm_->kNoError,
               apm_->echo_cancellation()->GetMetrics(&echo_metrics));
+    int median = 0;
+    int std = 0;
+    EXPECT_EQ(apm_->kNoError,
+              apm_->echo_cancellation()->GetDelayMetrics(&median, &std));
 #endif
 
     if (!write_output_data) {
@@ -612,6 +618,11 @@ TEST_F(ApmTest, Process) {
                 reference.echo_return_loss_enhancement());
       TestStats(echo_metrics.a_nlp,
                 reference.a_nlp());
+
+      webrtc::audioproc::Test::DelayMetrics reference_delay =
+          test->delay_metrics();
+      EXPECT_EQ(median, reference_delay.median());
+      EXPECT_EQ(std, reference_delay.std());
 #endif
     } else {
       test->set_has_echo_count(has_echo_count);
@@ -632,6 +643,11 @@ TEST_F(ApmTest, Process) {
                         message->mutable_echo_return_loss_enhancement());
       WriteStatsMessage(echo_metrics.a_nlp,
                         message->mutable_a_nlp());
+
+      webrtc::audioproc::Test::DelayMetrics* message_delay =
+          test->mutable_delay_metrics();
+      message_delay->set_median(median);
+      message_delay->set_std(std);
 #endif
     }
 
@@ -696,6 +712,18 @@ TEST_F(ApmTest, EchoCancellation) {
             apm_->echo_cancellation()->enable_metrics(false));
   EXPECT_FALSE(apm_->echo_cancellation()->are_metrics_enabled());
 
+  int median = 0;
+  int std = 0;
+  EXPECT_EQ(apm_->kNotEnabledError,
+            apm_->echo_cancellation()->GetDelayMetrics(&median, &std));
+
+  EXPECT_EQ(apm_->kNoError,
+            apm_->echo_cancellation()->enable_delay_logging(true));
+  EXPECT_TRUE(apm_->echo_cancellation()->is_delay_logging_enabled());
+  EXPECT_EQ(apm_->kNoError,
+            apm_->echo_cancellation()->enable_delay_logging(false));
+  EXPECT_FALSE(apm_->echo_cancellation()->is_delay_logging_enabled());
+
   EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(true));
   EXPECT_TRUE(apm_->echo_cancellation()->is_enabled());
   EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(false));

src/modules/audio_processing/main/test/unit_test/unittest.proto

@@ -35,6 +35,13 @@ message Test {
   }
 
   optional EchoMetrics echo_metrics = 11;
+
+  message DelayMetrics {
+    optional int32 median = 1;
+    optional int32 std = 2;
+  }
+
+  optional DelayMetrics delay_metrics = 12;
 }
 
 message OutputData {