diff --git a/webrtc/modules/audio_processing/BUILD.gn b/webrtc/modules/audio_processing/BUILD.gn
index 54c33e346c..b4d875934b 100644
--- a/webrtc/modules/audio_processing/BUILD.gn
+++ b/webrtc/modules/audio_processing/BUILD.gn
@@ -22,8 +22,6 @@ rtc_static_library("audio_processing") {
     "aec/aec_core.cc",
     "aec/aec_core.h",
     "aec/aec_core_optimized_methods.h",
-    "aec/aec_rdft.cc",
-    "aec/aec_rdft.h",
     "aec/aec_resampler.cc",
     "aec/aec_resampler.h",
     "aec/echo_cancellation.cc",
@@ -129,6 +127,9 @@ rtc_static_library("audio_processing") {
     "utility/delay_estimator_internal.h",
     "utility/delay_estimator_wrapper.cc",
     "utility/delay_estimator_wrapper.h",
+    "utility/ooura_fft.cc",
+    "utility/ooura_fft.h",
+    "utility/ooura_fft_tables_common.h",
     "vad/common.h",
     "vad/gmm.cc",
     "vad/gmm.h",
@@ -225,7 +226,7 @@ rtc_static_library("audio_processing") {
     if (mips_float_abi == "hard") {
       sources += [
         "aec/aec_core_mips.cc",
-        "aec/aec_rdft_mips.cc",
+        "utility/ooura_fft_mips.cc",
       ]
     }
   } else {
@@ -256,7 +257,8 @@ if (current_cpu == "x86" || current_cpu == "x64") {
   rtc_static_library("audio_processing_sse2") {
     sources = [
       "aec/aec_core_sse2.cc",
-      "aec/aec_rdft_sse2.cc",
+      "utility/ooura_fft_sse2.cc",
+      "utility/ooura_fft_tables_neon_sse2.h",
     ]
 
     if (is_posix) {
@@ -275,9 +277,10 @@ if (rtc_build_with_neon) {
   rtc_static_library("audio_processing_neon") {
     sources = [
       "aec/aec_core_neon.cc",
-      "aec/aec_rdft_neon.cc",
       "aecm/aecm_core_neon.cc",
       "ns/nsx_core_neon.c",
+      "utility/ooura_fft_neon.cc",
+      "utility/ooura_fft_tables_neon_sse2.h",
     ]
 
     if (current_cpu != "arm64") {
diff --git a/webrtc/modules/audio_processing/aec/aec_core.cc b/webrtc/modules/audio_processing/aec/aec_core.cc
index f249833269..e3fd14c9da 100644
--- a/webrtc/modules/audio_processing/aec/aec_core.cc
+++ b/webrtc/modules/audio_processing/aec/aec_core.cc
@@ -28,7 +28,6 @@ extern "C" {
 #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
 #include "webrtc/modules/audio_processing/aec/aec_common.h"
 #include "webrtc/modules/audio_processing/aec/aec_core_optimized_methods.h"
-#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
 #include "webrtc/modules/audio_processing/logging/apm_data_dumper.h"
 #include "webrtc/modules/audio_processing/utility/delay_estimator_wrapper.h"
 #include "webrtc/system_wrappers/include/cpu_features_wrapper.h"
@@ -337,6 +336,7 @@ static void ScaleErrorSignal(float mu,
 }
 
 static void FilterAdaptation(
+    const OouraFft& ooura_fft,
     int num_partitions,
     int x_fft_buf_block_pos,
     float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
@@ -364,7 +364,7 @@ static void FilterAdaptation(
         MulRe(x_fft_buf[0][xPos + PART_LEN], -x_fft_buf[1][xPos + PART_LEN],
               e_fft[0][PART_LEN], e_fft[1][PART_LEN]);
 
-    aec_rdft_inverse_128(fft);
+    ooura_fft.InverseFft(fft);
     memset(fft + PART_LEN, 0, sizeof(float) * PART_LEN);
 
     // fft scaling
@@ -374,7 +374,7 @@ static void FilterAdaptation(
         fft[j] *= scale;
       }
     }
-    aec_rdft_forward_128(fft);
+    ooura_fft.Fft(fft);
 
     h_fft_buf[0][pos] += fft[0];
     h_fft_buf[0][pos + PART_LEN] += fft[1];
@@ -835,7 +835,8 @@ static void UpdateDelayMetrics(AecCore* self) {
   return;
 }
 
-static void ScaledInverseFft(float freq_data[2][PART_LEN1],
+static void ScaledInverseFft(const OouraFft& ooura_fft,
+                             float freq_data[2][PART_LEN1],
                              float time_data[PART_LEN2],
                              float scale,
                              int conjugate) {
@@ -848,12 +849,14 @@ static void ScaledInverseFft(float freq_data[2][PART_LEN1],
     time_data[2 * i] = freq_data[0][i] * normalization;
     time_data[2 * i + 1] = sign * freq_data[1][i] * normalization;
   }
-  aec_rdft_inverse_128(time_data);
+  ooura_fft.InverseFft(time_data);
 }
 
-static void Fft(float time_data[PART_LEN2], float freq_data[2][PART_LEN1]) {
+static void Fft(const OouraFft& ooura_fft,
+                float time_data[PART_LEN2],
+                float freq_data[2][PART_LEN1]) {
   int i;
-  aec_rdft_forward_128(time_data);
+  ooura_fft.Fft(time_data);
 
   // Reorder fft output data.
   freq_data[1][0] = 0;
@@ -970,7 +973,8 @@ static void RegressorPower(int num_partitions,
   }
 }
 
-static void EchoSubtraction(int num_partitions,
+static void EchoSubtraction(const OouraFft& ooura_fft,
+                            int num_partitions,
                             int extended_filter_enabled,
                             int* extreme_filter_divergence,
                             float filter_step_size,
@@ -1019,7 +1023,7 @@ static void EchoSubtraction(int num_partitions,
                       h_fft_buf, s_fft);
 
   // Compute the time-domain echo estimate s.
-  ScaledInverseFft(s_fft, s_extended, 2.0f, 0);
+  ScaledInverseFft(ooura_fft, s_fft, s_extended, 2.0f, 0);
   s = &s_extended[PART_LEN];
 
   // Compute the time-domain echo prediction error.
@@ -1030,12 +1034,12 @@ static void EchoSubtraction(int num_partitions,
   // Compute the frequency domain echo prediction error.
   memset(e_extended, 0, sizeof(float) * PART_LEN);
   memcpy(e_extended + PART_LEN, e, sizeof(float) * PART_LEN);
-  Fft(e_extended, e_fft);
+  Fft(ooura_fft, e_extended, e_fft);
 
   // Scale error signal inversely with far power.
   WebRtcAec_ScaleErrorSignal(filter_step_size, error_threshold, x_pow, e_fft);
-  WebRtcAec_FilterAdaptation(num_partitions, *x_fft_buf_block_pos, x_fft_buf,
-                             e_fft, h_fft_buf);
+  WebRtcAec_FilterAdaptation(ooura_fft, num_partitions, *x_fft_buf_block_pos,
+                             x_fft_buf, e_fft, h_fft_buf);
   memcpy(echo_subtractor_output, e, sizeof(float) * PART_LEN);
 }
 
@@ -1152,7 +1156,8 @@ static void FormSuppressionGain(AecCore* aec,
   WebRtcAec_Overdrive(aec->overdrive_scaling, hNlFb, hNl);
 }
 
-static void EchoSuppression(AecCore* aec,
+static void EchoSuppression(const OouraFft& ooura_fft,
+                            AecCore* aec,
                             float* nearend_extended_block_lowest_band,
                             float farend_extended_block[PART_LEN2],
                             float* echo_subtractor_output,
@@ -1182,19 +1187,19 @@ static void EchoSuppression(AecCore* aec,
   // Analysis filter banks for the echo suppressor.
   // Windowed near-end ffts.
   WindowData(fft, nearend_extended_block_lowest_band);
-  aec_rdft_forward_128(fft);
+  ooura_fft.Fft(fft);
   StoreAsComplex(fft, dfw);
 
   // Windowed echo suppressor output ffts.
   WindowData(fft, aec->eBuf);
-  aec_rdft_forward_128(fft);
+  ooura_fft.Fft(fft);
   StoreAsComplex(fft, efw);
 
   // NLP
 
   // Convert far-end partition to the frequency domain with windowing.
   WindowData(fft, farend_extended_block);
-  Fft(fft, xfw);
+  Fft(ooura_fft, fft, xfw);
   xfw_ptr = &xfw[0][0];
 
   // Buffer far.
@@ -1236,7 +1241,7 @@ static void EchoSuppression(AecCore* aec,
                aec->noisePow, hNl);
 
   // Inverse error fft.
-  ScaledInverseFft(efw, fft, 2.0f, 1);
+  ScaledInverseFft(ooura_fft, efw, fft, 2.0f, 1);
 
   // Overlap and add to obtain output.
   for (i = 0; i < PART_LEN; i++) {
@@ -1257,7 +1262,7 @@ static void EchoSuppression(AecCore* aec,
     GetHighbandGain(hNl, &nlpGainHband);
 
     // Inverse comfort_noise
-    ScaledInverseFft(comfortNoiseHband, fft, 2.0f, 0);
+    ScaledInverseFft(ooura_fft, comfortNoiseHband, fft, 2.0f, 0);
 
     // compute gain factor
     for (j = 1; j < aec->num_bands; ++j) {
@@ -1330,7 +1335,7 @@ static void ProcessNearendBlock(
 
   // Convert far-end signal to the frequency domain.
   memcpy(fft, farend_extended_block_lowest_band, sizeof(float) * PART_LEN2);
-  Fft(fft, farend_fft);
+  Fft(aec->ooura_fft, fft, farend_fft);
 
   // Form extended nearend frame.
   memcpy(&nearend_extended_block_lowest_band[0],
@@ -1340,7 +1345,7 @@ static void ProcessNearendBlock(
 
   // Convert near-end signal to the frequency domain.
   memcpy(fft, nearend_extended_block_lowest_band, sizeof(float) * PART_LEN2);
-  Fft(fft, nearend_fft);
+  Fft(aec->ooura_fft, fft, nearend_fft);
 
   // Power smoothing.
   if (aec->refined_adaptive_filter_enabled) {
@@ -1419,11 +1424,11 @@ static void ProcessNearendBlock(
   }
 
   // Perform echo subtraction.
-  EchoSubtraction(aec->num_partitions, aec->extended_filter_enabled,
-                  &aec->extreme_filter_divergence, aec->filter_step_size,
-                  aec->error_threshold, &farend_fft[0][0], &aec->xfBufBlockPos,
-                  aec->xfBuf, &nearend_block[0][0], aec->xPow, aec->wfBuf,
-                  echo_subtractor_output);
+  EchoSubtraction(
+      aec->ooura_fft, aec->num_partitions, aec->extended_filter_enabled,
+      &aec->extreme_filter_divergence, aec->filter_step_size,
+      aec->error_threshold, &farend_fft[0][0], &aec->xfBufBlockPos, aec->xfBuf,
+      &nearend_block[0][0], aec->xPow, aec->wfBuf, echo_subtractor_output);
   aec->data_dumper->DumpRaw("aec_h_fft", PART_LEN1 * aec->num_partitions,
                             &aec->wfBuf[0][0]);
   aec->data_dumper->DumpRaw("aec_h_fft", PART_LEN1 * aec->num_partitions,
@@ -1438,7 +1443,7 @@ static void ProcessNearendBlock(
   }
 
   // Perform echo suppression.
-  EchoSuppression(aec, nearend_extended_block_lowest_band,
+  EchoSuppression(aec->ooura_fft, aec, nearend_extended_block_lowest_band,
                   farend_extended_block_lowest_band, echo_subtractor_output,
                   output_block);
 
@@ -1524,8 +1529,6 @@ AecCore* WebRtcAec_CreateAec(int instance_count) {
   WebRtcAec_InitAec_neon();
 #endif
 
-  aec_rdft_init();
-
   return aec;
 }
 
diff --git a/webrtc/modules/audio_processing/aec/aec_core.h b/webrtc/modules/audio_processing/aec/aec_core.h
index e41842efed..0c491980c2 100644
--- a/webrtc/modules/audio_processing/aec/aec_core.h
+++ b/webrtc/modules/audio_processing/aec/aec_core.h
@@ -26,6 +26,7 @@ extern "C" {
 #include "webrtc/common_audio/wav_file.h"
 #include "webrtc/modules/audio_processing/aec/aec_common.h"
 #include "webrtc/modules/audio_processing/utility/block_mean_calculator.h"
+#include "webrtc/modules/audio_processing/utility/ooura_fft.h"
 #include "webrtc/typedefs.h"
 
 namespace webrtc {
@@ -134,6 +135,7 @@ struct AecCore {
   ~AecCore();
 
   std::unique_ptr<ApmDataDumper> data_dumper;
+  const OouraFft ooura_fft;
 
   CoherenceState coherence_state;
 
diff --git a/webrtc/modules/audio_processing/aec/aec_core_mips.cc b/webrtc/modules/audio_processing/aec/aec_core_mips.cc
index a9b5cd4e60..93f075b830 100644
--- a/webrtc/modules/audio_processing/aec/aec_core_mips.cc
+++ b/webrtc/modules/audio_processing/aec/aec_core_mips.cc
@@ -20,7 +20,7 @@ extern "C" {
 #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
 }
 #include "webrtc/modules/audio_processing/aec/aec_core_optimized_methods.h"
-#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
+#include "webrtc/modules/audio_processing/utility/ooura_fft.h"
 
 namespace webrtc {
 
@@ -144,6 +144,7 @@ void WebRtcAec_FilterFar_mips(
 }
 
 void WebRtcAec_FilterAdaptation_mips(
+    const OouraFft& ooura_fft,
     int num_partitions,
     int x_fft_buf_block_pos,
     float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
@@ -238,7 +239,7 @@ void WebRtcAec_FilterAdaptation_mips(
       : [fft] "r" (fft)
       : "memory");
 
-    aec_rdft_inverse_128(fft);
+    ooura_fft.InverseFft(fft);
     memset(fft + PART_LEN, 0, sizeof(float) * PART_LEN);
 
     // fft scaling
@@ -285,7 +286,7 @@ void WebRtcAec_FilterAdaptation_mips(
         : [scale] "f" (scale), [fft] "r" (fft)
         : "memory");
     }
-    aec_rdft_forward_128(fft);
+    ooura_fft.Fft(fft);
     aRe = h_fft_buf[0] + pos;
     aIm = h_fft_buf[1] + pos;
     __asm __volatile(
diff --git a/webrtc/modules/audio_processing/aec/aec_core_neon.cc b/webrtc/modules/audio_processing/aec/aec_core_neon.cc
index bc503ba3db..4995ebf5d3 100644
--- a/webrtc/modules/audio_processing/aec/aec_core_neon.cc
+++ b/webrtc/modules/audio_processing/aec/aec_core_neon.cc
@@ -23,7 +23,7 @@ extern "C" {
 }
 #include "webrtc/modules/audio_processing/aec/aec_common.h"
 #include "webrtc/modules/audio_processing/aec/aec_core_optimized_methods.h"
-#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
+#include "webrtc/modules/audio_processing/utility/ooura_fft.h"
 
 namespace webrtc {
 
@@ -184,6 +184,7 @@ static void ScaleErrorSignalNEON(float mu,
 }
 
 static void FilterAdaptationNEON(
+    const OouraFft& ooura_fft,
     int num_partitions,
     int x_fft_buf_block_pos,
     float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
@@ -225,7 +226,7 @@ static void FilterAdaptationNEON(
         MulRe(x_fft_buf[0][xPos + PART_LEN], -x_fft_buf[1][xPos + PART_LEN],
               e_fft[0][PART_LEN], e_fft[1][PART_LEN]);
 
-    aec_rdft_inverse_128(fft);
+    ooura_fft.InverseFft(fft);
     memset(fft + PART_LEN, 0, sizeof(float) * PART_LEN);
 
     // fft scaling
@@ -238,7 +239,7 @@ static void FilterAdaptationNEON(
         vst1q_f32(&fft[j], fft_scale);
       }
     }
-    aec_rdft_forward_128(fft);
+    ooura_fft.Fft(fft);
 
     {
       const float wt1 = h_fft_buf[1][pos];
diff --git a/webrtc/modules/audio_processing/aec/aec_core_optimized_methods.h b/webrtc/modules/audio_processing/aec/aec_core_optimized_methods.h
index d1fb6e892a..5e873c8c80 100644
--- a/webrtc/modules/audio_processing/aec/aec_core_optimized_methods.h
+++ b/webrtc/modules/audio_processing/aec/aec_core_optimized_methods.h
@@ -31,6 +31,7 @@ typedef void (*WebRtcAecScaleErrorSignal)(float mu,
                                           float ef[2][PART_LEN1]);
 extern WebRtcAecScaleErrorSignal WebRtcAec_ScaleErrorSignal;
 typedef void (*WebRtcAecFilterAdaptation)(
+    const OouraFft& ooura_fft,
     int num_partitions,
     int x_fft_buf_block_pos,
     float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
diff --git a/webrtc/modules/audio_processing/aec/aec_core_sse2.cc b/webrtc/modules/audio_processing/aec/aec_core_sse2.cc
index 47ba12f419..ac93919a77 100644
--- a/webrtc/modules/audio_processing/aec/aec_core_sse2.cc
+++ b/webrtc/modules/audio_processing/aec/aec_core_sse2.cc
@@ -21,7 +21,7 @@ extern "C" {
 }
 #include "webrtc/modules/audio_processing/aec/aec_common.h"
 #include "webrtc/modules/audio_processing/aec/aec_core_optimized_methods.h"
-#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
+#include "webrtc/modules/audio_processing/utility/ooura_fft.h"
 
 namespace webrtc {
 
@@ -140,6 +140,7 @@ static void ScaleErrorSignalSSE2(float mu,
 }
 
 static void FilterAdaptationSSE2(
+    const OouraFft& ooura_fft,
     int num_partitions,
     int x_fft_buf_block_pos,
     float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
@@ -183,7 +184,7 @@ static void FilterAdaptationSSE2(
         MulRe(x_fft_buf[0][xPos + PART_LEN], -x_fft_buf[1][xPos + PART_LEN],
               e_fft[0][PART_LEN], e_fft[1][PART_LEN]);
 
-    aec_rdft_inverse_128(fft);
+    ooura_fft.InverseFft(fft);
     memset(fft + PART_LEN, 0, sizeof(float) * PART_LEN);
 
     // fft scaling
@@ -196,7 +197,7 @@ static void FilterAdaptationSSE2(
         _mm_storeu_ps(&fft[j], fft_scale);
       }
     }
-    aec_rdft_forward_128(fft);
+    ooura_fft.Fft(fft);
 
     {
       float wt1 = h_fft_buf[1][pos];
diff --git a/webrtc/modules/audio_processing/aec/aec_rdft.h b/webrtc/modules/audio_processing/aec/aec_rdft.h
deleted file mode 100644
index d83eb27d83..0000000000
--- a/webrtc/modules/audio_processing/aec/aec_rdft.h
+++ /dev/null
@@ -1,61 +0,0 @@
-/*
- *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_RDFT_H_
-#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_RDFT_H_
-
-#include "webrtc/modules/audio_processing/aec/aec_common.h"
-
-// These intrinsics were unavailable before VS 2008.
-// TODO(andrew): move to a common file.
-#if defined(_MSC_VER) && _MSC_VER < 1500
-#include <emmintrin.h>
-static __inline __m128 _mm_castsi128_ps(__m128i a) { return *(__m128*)&a; }
-static __inline __m128i _mm_castps_si128(__m128 a) { return *(__m128i*)&a; }
-#endif
-
-// Constants shared by all paths (C, SSE2, NEON).
-extern const float rdft_w[64];
-// Constants used by the C path.
-extern const float rdft_wk3ri_first[16];
-extern const float rdft_wk3ri_second[16];
-// Constants used by SSE2 and NEON but initialized in the C path.
-extern ALIGN16_BEG const float ALIGN16_END rdft_wk1r[32];
-extern ALIGN16_BEG const float ALIGN16_END rdft_wk2r[32];
-extern ALIGN16_BEG const float ALIGN16_END rdft_wk3r[32];
-extern ALIGN16_BEG const float ALIGN16_END rdft_wk1i[32];
-extern ALIGN16_BEG const float ALIGN16_END rdft_wk2i[32];
-extern ALIGN16_BEG const float ALIGN16_END rdft_wk3i[32];
-extern ALIGN16_BEG const float ALIGN16_END cftmdl_wk1r[4];
-
-// code path selection function pointers
-typedef void (*RftSub128)(float* a);
-extern RftSub128 rftfsub_128;
-extern RftSub128 rftbsub_128;
-extern RftSub128 cft1st_128;
-extern RftSub128 cftmdl_128;
-extern RftSub128 cftfsub_128;
-extern RftSub128 cftbsub_128;
-extern RftSub128 bitrv2_128;
-
-// entry points
-void aec_rdft_init(void);
-void aec_rdft_init_sse2(void);
-void aec_rdft_forward_128(float* a);
-void aec_rdft_inverse_128(float* a);
-
-#if defined(MIPS_FPU_LE)
-void aec_rdft_init_mips(void);
-#endif
-#if defined(WEBRTC_HAS_NEON)
-void aec_rdft_init_neon(void);
-#endif
-
-#endif  // WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_RDFT_H_
diff --git a/webrtc/modules/audio_processing/audio_processing.gypi b/webrtc/modules/audio_processing/audio_processing.gypi
index 78f133a97a..93387891eb 100644
--- a/webrtc/modules/audio_processing/audio_processing.gypi
+++ b/webrtc/modules/audio_processing/audio_processing.gypi
@@ -34,8 +34,6 @@
         'aec/aec_core.cc',
         'aec/aec_core.h',
         'aec/aec_core_optimized_methods.h',
-        'aec/aec_rdft.cc',
-        'aec/aec_rdft.h',
         'aec/aec_resampler.cc',
         'aec/aec_resampler.h',
         'aec/echo_cancellation.cc',
@@ -141,6 +139,9 @@
         'utility/delay_estimator_internal.h',
         'utility/delay_estimator_wrapper.cc',
         'utility/delay_estimator_wrapper.h',
+        'utility/ooura_fft.cc',
+        'utility/ooura_fft.h',
+        'utility/ooura_fft_tables_common.h',
         'vad/common.h',
         'vad/gmm.cc',
         'vad/gmm.h',
@@ -236,7 +237,7 @@
             ['mips_float_abi=="hard"', {
               'sources': [
                 'aec/aec_core_mips.cc',
-                'aec/aec_rdft_mips.cc',
+                'utility/ooura_fft_mips.cc',
               ],
             }],
           ],
@@ -275,7 +276,8 @@
           'type': 'static_library',
           'sources': [
             'aec/aec_core_sse2.cc',
-            'aec/aec_rdft_sse2.cc',
+            'utility/ooura_fft_sse2.cc',
+            'utility/ooura_fft_tables_neon_sse2.h',
           ],
           'conditions': [
             ['apm_debug_dump==1', {
@@ -303,9 +305,10 @@
         ],
         'sources': [
           'aec/aec_core_neon.cc',
-          'aec/aec_rdft_neon.cc',
           'aecm/aecm_core_neon.cc',
           'ns/nsx_core_neon.c',
+          'utility/ooura_fft_neon.cc',
+          'utility/ooura_fft_tables_neon_sse2.h',
         ],
         'conditions': [
           ['apm_debug_dump==1', {
diff --git a/webrtc/modules/audio_processing/level_controller/signal_classifier.cc b/webrtc/modules/audio_processing/level_controller/signal_classifier.cc
index c9a07e754d..dd67737403 100644
--- a/webrtc/modules/audio_processing/level_controller/signal_classifier.cc
+++ b/webrtc/modules/audio_processing/level_controller/signal_classifier.cc
@@ -16,7 +16,6 @@
 
 #include "webrtc/base/array_view.h"
 #include "webrtc/base/constructormagic.h"
-#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
 #include "webrtc/modules/audio_processing/audio_buffer.h"
 #include "webrtc/modules/audio_processing/level_controller/down_sampler.h"
 #include "webrtc/modules/audio_processing/level_controller/noise_spectrum_estimator.h"
@@ -35,13 +34,14 @@ void RemoveDcLevel(rtc::ArrayView<float> x) {
   }
 }
 
-void PowerSpectrum(rtc::ArrayView<const float> x,
+void PowerSpectrum(const OouraFft* ooura_fft,
+                   rtc::ArrayView<const float> x,
                    rtc::ArrayView<float> spectrum) {
   RTC_DCHECK_EQ(65u, spectrum.size());
   RTC_DCHECK_EQ(128u, x.size());
   float X[128];
   std::copy(x.data(), x.data() + x.size(), X);
-  aec_rdft_forward_128(X);
+  ooura_fft->Fft(X);
 
   float* X_p = X;
   RTC_DCHECK_EQ(X_p, &X[0]);
@@ -118,7 +118,6 @@ SignalClassifier::SignalClassifier(ApmDataDumper* data_dumper)
 SignalClassifier::~SignalClassifier() {}
 
 void SignalClassifier::Initialize(int sample_rate_hz) {
-  aec_rdft_init();
   down_sampler_.Initialize(sample_rate_hz);
   noise_spectrum_estimator_.Initialize();
   frame_extender_.reset(new FrameExtender(80, 128));
@@ -141,7 +140,7 @@ void SignalClassifier::Analyze(const AudioBuffer& audio,
   frame_extender_->ExtendFrame(downsampled_frame, extended_frame);
   RemoveDcLevel(extended_frame);
   float signal_spectrum[65];
-  PowerSpectrum(extended_frame, signal_spectrum);
+  PowerSpectrum(&ooura_fft_, extended_frame, signal_spectrum);
 
   // Classify the signal based on the estimate of the noise spectrum and the
   // signal spectrum estimate.
diff --git a/webrtc/modules/audio_processing/level_controller/signal_classifier.h b/webrtc/modules/audio_processing/level_controller/signal_classifier.h
index 2afa724c56..8c791fec7e 100644
--- a/webrtc/modules/audio_processing/level_controller/signal_classifier.h
+++ b/webrtc/modules/audio_processing/level_controller/signal_classifier.h
@@ -18,6 +18,7 @@
 #include "webrtc/base/constructormagic.h"
 #include "webrtc/modules/audio_processing/level_controller/down_sampler.h"
 #include "webrtc/modules/audio_processing/level_controller/noise_spectrum_estimator.h"
+#include "webrtc/modules/audio_processing/utility/ooura_fft.h"
 
 namespace webrtc {
 
@@ -57,6 +58,7 @@ class SignalClassifier {
   int initialization_frames_left_;
   int consistent_classification_counter_;
   SignalType last_signal_type_;
+  const OouraFft ooura_fft_;
   RTC_DISALLOW_IMPLICIT_CONSTRUCTORS(SignalClassifier);
 };
 
diff --git a/webrtc/modules/audio_processing/aec/aec_rdft.cc b/webrtc/modules/audio_processing/utility/ooura_fft.cc
similarity index 65%
rename from webrtc/modules/audio_processing/aec/aec_rdft.cc
rename to webrtc/modules/audio_processing/utility/ooura_fft.cc
index 690fe9f34f..4ba88d7f17 100644
--- a/webrtc/modules/audio_processing/aec/aec_rdft.cc
+++ b/webrtc/modules/audio_processing/utility/ooura_fft.cc
@@ -10,6 +10,8 @@
  *    - Trivial type modifications.
  *    - Minimal code subset to do rdft of length 128.
  *    - Optimizations because of known length.
+ *    - Removed the global variables by moving the code in to a class in order
+ *      to make it thread safe.
  *
  *  All changes are covered by the WebRTC license and IP grant:
  *  Use of this source code is governed by a BSD-style license
@@ -19,184 +21,19 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
+#include "webrtc/modules/audio_processing//utility/ooura_fft.h"
 
 #include <math.h>
 
+#include "webrtc/modules/audio_processing/utility/ooura_fft_tables_common.h"
 #include "webrtc/system_wrappers/include/cpu_features_wrapper.h"
 #include "webrtc/typedefs.h"
 
-// These tables used to be computed at run-time. For example, refer to:
-// https://code.google.com/p/webrtc/source/browse/trunk/webrtc/modules/audio_processing/aec/aec_rdft.c?r=6564
-// to see the initialization code.
-const float rdft_w[64] = {
-    1.0000000000f, 0.0000000000f, 0.7071067691f, 0.7071067691f,
-    0.9238795638f, 0.3826834559f, 0.3826834559f, 0.9238795638f,
-    0.9807852507f, 0.1950903237f, 0.5555702448f, 0.8314695954f,
-    0.8314695954f, 0.5555702448f, 0.1950903237f, 0.9807852507f,
-    0.9951847196f, 0.0980171412f, 0.6343933344f, 0.7730104327f,
-    0.8819212914f, 0.4713967443f, 0.2902846634f, 0.9569403529f,
-    0.9569403529f, 0.2902846634f, 0.4713967443f, 0.8819212914f,
-    0.7730104327f, 0.6343933344f, 0.0980171412f, 0.9951847196f,
-    0.7071067691f, 0.4993977249f, 0.4975923598f, 0.4945882559f,
-    0.4903926253f, 0.4850156307f, 0.4784701765f, 0.4707720280f,
-    0.4619397819f, 0.4519946277f, 0.4409606457f, 0.4288643003f,
-    0.4157347977f, 0.4016037583f, 0.3865052164f, 0.3704755902f,
-    0.3535533845f, 0.3357794881f, 0.3171966672f, 0.2978496552f,
-    0.2777851224f, 0.2570513785f, 0.2356983721f, 0.2137775421f,
-    0.1913417280f, 0.1684449315f, 0.1451423317f, 0.1214900985f,
-    0.0975451618f, 0.0733652338f, 0.0490085706f, 0.0245338380f,
-};
-const float rdft_wk3ri_first[16] = {
-    1.000000000f, 0.000000000f, 0.382683456f, 0.923879564f,
-    0.831469536f, 0.555570245f, -0.195090353f, 0.980785251f,
-    0.956940353f, 0.290284693f, 0.098017156f, 0.995184720f,
-    0.634393334f, 0.773010492f, -0.471396863f, 0.881921172f,
-};
-const float rdft_wk3ri_second[16] = {
-    -0.707106769f, 0.707106769f, -0.923879564f, -0.382683456f,
-    -0.980785251f, 0.195090353f, -0.555570245f, -0.831469536f,
-    -0.881921172f, 0.471396863f, -0.773010492f, -0.634393334f,
-    -0.995184720f, -0.098017156f, -0.290284693f, -0.956940353f,
-};
-ALIGN16_BEG const float ALIGN16_END rdft_wk1r[32] = {
-    1.000000000f, 1.000000000f, 0.707106769f, 0.707106769f,
-    0.923879564f, 0.923879564f, 0.382683456f, 0.382683456f,
-    0.980785251f, 0.980785251f, 0.555570245f, 0.555570245f,
-    0.831469595f, 0.831469595f, 0.195090324f, 0.195090324f,
-    0.995184720f, 0.995184720f, 0.634393334f, 0.634393334f,
-    0.881921291f, 0.881921291f, 0.290284663f, 0.290284663f,
-    0.956940353f, 0.956940353f, 0.471396744f, 0.471396744f,
-    0.773010433f, 0.773010433f, 0.098017141f, 0.098017141f,
-};
-ALIGN16_BEG const float ALIGN16_END rdft_wk2r[32] = {
-    1.000000000f, 1.000000000f, -0.000000000f, -0.000000000f,
-    0.707106769f, 0.707106769f, -0.707106769f, -0.707106769f,
-    0.923879564f, 0.923879564f, -0.382683456f, -0.382683456f,
-    0.382683456f, 0.382683456f, -0.923879564f, -0.923879564f,
-    0.980785251f, 0.980785251f, -0.195090324f, -0.195090324f,
-    0.555570245f, 0.555570245f, -0.831469595f, -0.831469595f,
-    0.831469595f, 0.831469595f, -0.555570245f, -0.555570245f,
-    0.195090324f, 0.195090324f, -0.980785251f, -0.980785251f,
-};
-ALIGN16_BEG const float ALIGN16_END rdft_wk3r[32] = {
-    1.000000000f, 1.000000000f, -0.707106769f, -0.707106769f,
-    0.382683456f, 0.382683456f, -0.923879564f, -0.923879564f,
-    0.831469536f, 0.831469536f, -0.980785251f, -0.980785251f,
-    -0.195090353f, -0.195090353f, -0.555570245f, -0.555570245f,
-    0.956940353f, 0.956940353f, -0.881921172f, -0.881921172f,
-    0.098017156f, 0.098017156f, -0.773010492f, -0.773010492f,
-    0.634393334f, 0.634393334f, -0.995184720f, -0.995184720f,
-    -0.471396863f, -0.471396863f, -0.290284693f, -0.290284693f,
-};
-ALIGN16_BEG const float ALIGN16_END rdft_wk1i[32] = {
-    -0.000000000f, 0.000000000f, -0.707106769f, 0.707106769f,
-    -0.382683456f, 0.382683456f, -0.923879564f, 0.923879564f,
-    -0.195090324f, 0.195090324f, -0.831469595f, 0.831469595f,
-    -0.555570245f, 0.555570245f, -0.980785251f, 0.980785251f,
-    -0.098017141f, 0.098017141f, -0.773010433f, 0.773010433f,
-    -0.471396744f, 0.471396744f, -0.956940353f, 0.956940353f,
-    -0.290284663f, 0.290284663f, -0.881921291f, 0.881921291f,
-    -0.634393334f, 0.634393334f, -0.995184720f, 0.995184720f,
-};
-ALIGN16_BEG const float ALIGN16_END rdft_wk2i[32] = {
-    -0.000000000f, 0.000000000f, -1.000000000f, 1.000000000f,
-    -0.707106769f, 0.707106769f, -0.707106769f, 0.707106769f,
-    -0.382683456f, 0.382683456f, -0.923879564f, 0.923879564f,
-    -0.923879564f, 0.923879564f, -0.382683456f, 0.382683456f,
-    -0.195090324f, 0.195090324f, -0.980785251f, 0.980785251f,
-    -0.831469595f, 0.831469595f, -0.555570245f, 0.555570245f,
-    -0.555570245f, 0.555570245f, -0.831469595f, 0.831469595f,
-    -0.980785251f, 0.980785251f, -0.195090324f, 0.195090324f,
-};
-ALIGN16_BEG const float ALIGN16_END rdft_wk3i[32] = {
-    -0.000000000f, 0.000000000f, -0.707106769f, 0.707106769f,
-    -0.923879564f, 0.923879564f, 0.382683456f, -0.382683456f,
-    -0.555570245f, 0.555570245f, -0.195090353f, 0.195090353f,
-    -0.980785251f, 0.980785251f, 0.831469536f, -0.831469536f,
-    -0.290284693f, 0.290284693f, -0.471396863f, 0.471396863f,
-    -0.995184720f, 0.995184720f, 0.634393334f, -0.634393334f,
-    -0.773010492f, 0.773010492f, 0.098017156f, -0.098017156f,
-    -0.881921172f, 0.881921172f, 0.956940353f, -0.956940353f,
-};
-ALIGN16_BEG const float ALIGN16_END cftmdl_wk1r[4] = {
-    0.707106769f, 0.707106769f, 0.707106769f, -0.707106769f,
-};
+namespace webrtc {
 
-static void bitrv2_128_C(float* a) {
-  /*
-      Following things have been attempted but are no faster:
-      (a) Storing the swap indexes in a LUT (index calculations are done
-          for 'free' while waiting on memory/L1).
-      (b) Consolidate the load/store of two consecutive floats by a 64 bit
-          integer (execution is memory/L1 bound).
-      (c) Do a mix of floats and 64 bit integer to maximize register
-          utilization (execution is memory/L1 bound).
-      (d) Replacing ip[i] by ((k<<31)>>25) + ((k >> 1)<<5).
-      (e) Hard-coding of the offsets to completely eliminates index
-          calculations.
-  */
-
-  unsigned int j, j1, k, k1;
-  float xr, xi, yr, yi;
-
-  static const int ip[4] = {0, 64, 32, 96};
-  for (k = 0; k < 4; k++) {
-    for (j = 0; j < k; j++) {
-      j1 = 2 * j + ip[k];
-      k1 = 2 * k + ip[j];
-      xr = a[j1 + 0];
-      xi = a[j1 + 1];
-      yr = a[k1 + 0];
-      yi = a[k1 + 1];
-      a[j1 + 0] = yr;
-      a[j1 + 1] = yi;
-      a[k1 + 0] = xr;
-      a[k1 + 1] = xi;
-      j1 += 8;
-      k1 += 16;
-      xr = a[j1 + 0];
-      xi = a[j1 + 1];
-      yr = a[k1 + 0];
-      yi = a[k1 + 1];
-      a[j1 + 0] = yr;
-      a[j1 + 1] = yi;
-      a[k1 + 0] = xr;
-      a[k1 + 1] = xi;
-      j1 += 8;
-      k1 -= 8;
-      xr = a[j1 + 0];
-      xi = a[j1 + 1];
-      yr = a[k1 + 0];
-      yi = a[k1 + 1];
-      a[j1 + 0] = yr;
-      a[j1 + 1] = yi;
-      a[k1 + 0] = xr;
-      a[k1 + 1] = xi;
-      j1 += 8;
-      k1 += 16;
-      xr = a[j1 + 0];
-      xi = a[j1 + 1];
-      yr = a[k1 + 0];
-      yi = a[k1 + 1];
-      a[j1 + 0] = yr;
-      a[j1 + 1] = yi;
-      a[k1 + 0] = xr;
-      a[k1 + 1] = xi;
-    }
-    j1 = 2 * k + 8 + ip[k];
-    k1 = j1 + 8;
-    xr = a[j1 + 0];
-    xi = a[j1 + 1];
-    yr = a[k1 + 0];
-    yi = a[k1 + 1];
-    a[j1 + 0] = yr;
-    a[j1 + 1] = yi;
-    a[k1 + 0] = xr;
-    a[k1 + 1] = xi;
-  }
-}
+namespace {
 
+#if !(defined(MIPS_FPU_LE) || defined(WEBRTC_HAS_NEON))
 static void cft1st_128_C(float* a) {
   const int n = 128;
   int j, k1, k2;
@@ -431,67 +268,6 @@ static void cftmdl_128_C(float* a) {
   }
 }
 
-static void cftfsub_128_C(float* a) {
-  int j, j1, j2, j3, l;
-  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
-
-  cft1st_128(a);
-  cftmdl_128(a);
-  l = 32;
-  for (j = 0; j < l; j += 2) {
-    j1 = j + l;
-    j2 = j1 + l;
-    j3 = j2 + l;
-    x0r = a[j] + a[j1];
-    x0i = a[j + 1] + a[j1 + 1];
-    x1r = a[j] - a[j1];
-    x1i = a[j + 1] - a[j1 + 1];
-    x2r = a[j2] + a[j3];
-    x2i = a[j2 + 1] + a[j3 + 1];
-    x3r = a[j2] - a[j3];
-    x3i = a[j2 + 1] - a[j3 + 1];
-    a[j] = x0r + x2r;
-    a[j + 1] = x0i + x2i;
-    a[j2] = x0r - x2r;
-    a[j2 + 1] = x0i - x2i;
-    a[j1] = x1r - x3i;
-    a[j1 + 1] = x1i + x3r;
-    a[j3] = x1r + x3i;
-    a[j3 + 1] = x1i - x3r;
-  }
-}
-
-static void cftbsub_128_C(float* a) {
-  int j, j1, j2, j3, l;
-  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
-
-  cft1st_128(a);
-  cftmdl_128(a);
-  l = 32;
-
-  for (j = 0; j < l; j += 2) {
-    j1 = j + l;
-    j2 = j1 + l;
-    j3 = j2 + l;
-    x0r = a[j] + a[j1];
-    x0i = -a[j + 1] - a[j1 + 1];
-    x1r = a[j] - a[j1];
-    x1i = -a[j + 1] + a[j1 + 1];
-    x2r = a[j2] + a[j3];
-    x2i = a[j2 + 1] + a[j3 + 1];
-    x3r = a[j2] - a[j3];
-    x3i = a[j2 + 1] - a[j3 + 1];
-    a[j] = x0r + x2r;
-    a[j + 1] = x0i - x2i;
-    a[j2] = x0r - x2r;
-    a[j2 + 1] = x0i + x2i;
-    a[j1] = x1r - x3i;
-    a[j1 + 1] = x1i - x3r;
-    a[j3] = x1r + x3i;
-    a[j3 + 1] = x1i + x3r;
-  }
-}
-
 static void rftfsub_128_C(float* a) {
   const float* c = rdft_w + 32;
   int j1, j2, k1, k2;
@@ -535,8 +311,22 @@ static void rftbsub_128_C(float* a) {
   }
   a[65] = -a[65];
 }
+#endif
 
-void aec_rdft_forward_128(float* a) {
+
+}  // namespace
+
+OouraFft::OouraFft() {
+#if defined(WEBRTC_ARCH_X86_FAMILY)
+  use_sse2_ = (WebRtc_GetCPUInfo(kSSE2) != 0);
+#else
+  use_sse2_ = false;
+#endif
+}
+
+OouraFft::~OouraFft() = default;
+
+void OouraFft::Fft(float* a) const {
   float xi;
   bitrv2_128(a);
   cftfsub_128(a);
@@ -545,8 +335,7 @@ void aec_rdft_forward_128(float* a) {
   a[0] += a[1];
   a[1] = xi;
 }
-
-void aec_rdft_inverse_128(float* a) {
+void OouraFft::InverseFft(float* a) const {
   a[1] = 0.5f * (a[0] - a[1]);
   a[0] -= a[1];
   rftbsub_128(a);
@@ -554,32 +343,193 @@ void aec_rdft_inverse_128(float* a) {
   cftbsub_128(a);
 }
 
-// code path selection
-RftSub128 cft1st_128;
-RftSub128 cftmdl_128;
-RftSub128 rftfsub_128;
-RftSub128 rftbsub_128;
-RftSub128 cftfsub_128;
-RftSub128 cftbsub_128;
-RftSub128 bitrv2_128;
-
-void aec_rdft_init(void) {
-  cft1st_128 = cft1st_128_C;
-  cftmdl_128 = cftmdl_128_C;
-  rftfsub_128 = rftfsub_128_C;
-  rftbsub_128 = rftbsub_128_C;
-  cftfsub_128 = cftfsub_128_C;
-  cftbsub_128 = cftbsub_128_C;
-  bitrv2_128 = bitrv2_128_C;
-#if defined(WEBRTC_ARCH_X86_FAMILY)
-  if (WebRtc_GetCPUInfo(kSSE2)) {
-    aec_rdft_init_sse2();
+void OouraFft::cft1st_128(float* a) const {
+#if defined(MIPS_FPU_LE)
+  cft1st_128_mips(a);
+#elif defined(WEBRTC_HAS_NEON)
+  cft1st_128_neon(a);
+#else
+  if (use_sse2_) {
+    cft1st_128_SSE2(a);
+  } else {
+    cft1st_128_C(a);
   }
 #endif
+}
+void OouraFft::cftmdl_128(float* a) const {
 #if defined(MIPS_FPU_LE)
-  aec_rdft_init_mips();
-#endif
-#if defined(WEBRTC_HAS_NEON)
-  aec_rdft_init_neon();
+  cftmdl_128_mips(a);
+#elif defined(WEBRTC_HAS_NEON)
+  cftmdl_128_neon(a);
+#else
+  if (use_sse2_) {
+    cftmdl_128_SSE2(a);
+  } else {
+    cftmdl_128_C(a);
+  }
 #endif
 }
+void OouraFft::rftfsub_128(float* a) const {
+#if defined(MIPS_FPU_LE)
+  rftfsub_128_mips(a);
+#elif defined(WEBRTC_HAS_NEON)
+  rftfsub_128_neon(a);
+#else
+  if (use_sse2_) {
+    rftfsub_128_SSE2(a);
+  } else {
+    rftfsub_128_C(a);
+  }
+#endif
+}
+
+void OouraFft::rftbsub_128(float* a) const {
+#if defined(MIPS_FPU_LE)
+  rftbsub_128_mips(a);
+#elif defined(WEBRTC_HAS_NEON)
+  rftbsub_128_neon(a);
+#else
+  if (use_sse2_) {
+    rftbsub_128_SSE2(a);
+  } else {
+    rftbsub_128_C(a);
+  }
+#endif
+}
+
+void OouraFft::cftbsub_128(float* a) const {
+  int j, j1, j2, j3, l;
+  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
+
+  cft1st_128(a);
+  cftmdl_128(a);
+  l = 32;
+
+  for (j = 0; j < l; j += 2) {
+    j1 = j + l;
+    j2 = j1 + l;
+    j3 = j2 + l;
+    x0r = a[j] + a[j1];
+    x0i = -a[j + 1] - a[j1 + 1];
+    x1r = a[j] - a[j1];
+    x1i = -a[j + 1] + a[j1 + 1];
+    x2r = a[j2] + a[j3];
+    x2i = a[j2 + 1] + a[j3 + 1];
+    x3r = a[j2] - a[j3];
+    x3i = a[j2 + 1] - a[j3 + 1];
+    a[j] = x0r + x2r;
+    a[j + 1] = x0i - x2i;
+    a[j2] = x0r - x2r;
+    a[j2 + 1] = x0i + x2i;
+    a[j1] = x1r - x3i;
+    a[j1 + 1] = x1i - x3r;
+    a[j3] = x1r + x3i;
+    a[j3 + 1] = x1i + x3r;
+  }
+}
+
+void OouraFft::cftfsub_128(float* a) const {
+  int j, j1, j2, j3, l;
+  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
+
+  cft1st_128(a);
+  cftmdl_128(a);
+  l = 32;
+  for (j = 0; j < l; j += 2) {
+    j1 = j + l;
+    j2 = j1 + l;
+    j3 = j2 + l;
+    x0r = a[j] + a[j1];
+    x0i = a[j + 1] + a[j1 + 1];
+    x1r = a[j] - a[j1];
+    x1i = a[j + 1] - a[j1 + 1];
+    x2r = a[j2] + a[j3];
+    x2i = a[j2 + 1] + a[j3 + 1];
+    x3r = a[j2] - a[j3];
+    x3i = a[j2 + 1] - a[j3 + 1];
+    a[j] = x0r + x2r;
+    a[j + 1] = x0i + x2i;
+    a[j2] = x0r - x2r;
+    a[j2 + 1] = x0i - x2i;
+    a[j1] = x1r - x3i;
+    a[j1 + 1] = x1i + x3r;
+    a[j3] = x1r + x3i;
+    a[j3 + 1] = x1i - x3r;
+  }
+}
+
+void OouraFft::bitrv2_128(float* a) const {
+  /*
+      Following things have been attempted but are no faster:
+      (a) Storing the swap indexes in a LUT (index calculations are done
+          for 'free' while waiting on memory/L1).
+      (b) Consolidate the load/store of two consecutive floats by a 64 bit
+          integer (execution is memory/L1 bound).
+      (c) Do a mix of floats and 64 bit integer to maximize register
+          utilization (execution is memory/L1 bound).
+      (d) Replacing ip[i] by ((k<<31)>>25) + ((k >> 1)<<5).
+      (e) Hard-coding of the offsets to completely eliminates index
+          calculations.
+  */
+
+  unsigned int j, j1, k, k1;
+  float xr, xi, yr, yi;
+
+  const int ip[4] = {0, 64, 32, 96};
+  for (k = 0; k < 4; k++) {
+    for (j = 0; j < k; j++) {
+      j1 = 2 * j + ip[k];
+      k1 = 2 * k + ip[j];
+      xr = a[j1 + 0];
+      xi = a[j1 + 1];
+      yr = a[k1 + 0];
+      yi = a[k1 + 1];
+      a[j1 + 0] = yr;
+      a[j1 + 1] = yi;
+      a[k1 + 0] = xr;
+      a[k1 + 1] = xi;
+      j1 += 8;
+      k1 += 16;
+      xr = a[j1 + 0];
+      xi = a[j1 + 1];
+      yr = a[k1 + 0];
+      yi = a[k1 + 1];
+      a[j1 + 0] = yr;
+      a[j1 + 1] = yi;
+      a[k1 + 0] = xr;
+      a[k1 + 1] = xi;
+      j1 += 8;
+      k1 -= 8;
+      xr = a[j1 + 0];
+      xi = a[j1 + 1];
+      yr = a[k1 + 0];
+      yi = a[k1 + 1];
+      a[j1 + 0] = yr;
+      a[j1 + 1] = yi;
+      a[k1 + 0] = xr;
+      a[k1 + 1] = xi;
+      j1 += 8;
+      k1 += 16;
+      xr = a[j1 + 0];
+      xi = a[j1 + 1];
+      yr = a[k1 + 0];
+      yi = a[k1 + 1];
+      a[j1 + 0] = yr;
+      a[j1 + 1] = yi;
+      a[k1 + 0] = xr;
+      a[k1 + 1] = xi;
+    }
+    j1 = 2 * k + 8 + ip[k];
+    k1 = j1 + 8;
+    xr = a[j1 + 0];
+    xi = a[j1 + 1];
+    yr = a[k1 + 0];
+    yi = a[k1 + 1];
+    a[j1 + 0] = yr;
+    a[j1 + 1] = yi;
+    a[k1 + 0] = xr;
+    a[k1 + 1] = xi;
+  }
+}
+
+}  // namespace webrtc
diff --git a/webrtc/modules/audio_processing/utility/ooura_fft.h b/webrtc/modules/audio_processing/utility/ooura_fft.h
new file mode 100644
index 0000000000..a1b9f04f54
--- /dev/null
+++ b/webrtc/modules/audio_processing/utility/ooura_fft.h
@@ -0,0 +1,60 @@
+/*
+ *  Copyright (c) 2016 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_UTILITY_OOURA_FFT_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_UTILITY_OOURA_FFT_H_
+
+#include "webrtc/typedefs.h"
+
+namespace webrtc {
+
+#if defined(WEBRTC_ARCH_X86_FAMILY)
+void cft1st_128_SSE2(float* a);
+void cftmdl_128_SSE2(float* a);
+void rftfsub_128_SSE2(float* a);
+void rftbsub_128_SSE2(float* a);
+#endif
+
+#if defined(MIPS_FPU_LE)
+void cft1st_128_mips(float* a);
+void cftmdl_128_mips(float* a);
+void rftfsub_128_mips(float* a);
+void rftbsub_128_mips(float* a);
+#endif
+
+#if defined(WEBRTC_HAS_NEON)
+void cft1st_128_neon(float* a);
+void cftmdl_128_neon(float* a);
+void rftfsub_128_neon(float* a);
+void rftbsub_128_neon(float* a);
+#endif
+
+class OouraFft {
+ public:
+  OouraFft();
+  ~OouraFft();
+  void Fft(float* a) const;
+  void InverseFft(float* a) const;
+
+ private:
+  void cft1st_128(float* a) const;
+  void cftmdl_128(float* a) const;
+  void rftfsub_128(float* a) const;
+  void rftbsub_128(float* a) const;
+
+  void cftfsub_128(float* a) const;
+  void cftbsub_128(float* a) const;
+  void bitrv2_128(float* a) const;
+  bool use_sse2_;
+};
+
+}  // namespace webrtc
+
+#endif  // WEBRTC_MODULES_AUDIO_PROCESSING_UTILITY_OOURA_FFT_H_
diff --git a/webrtc/modules/audio_processing/aec/aec_rdft_mips.cc b/webrtc/modules/audio_processing/utility/ooura_fft_mips.cc
similarity index 98%
rename from webrtc/modules/audio_processing/aec/aec_rdft_mips.cc
rename to webrtc/modules/audio_processing/utility/ooura_fft_mips.cc
index 7e64e65716..3e9db6e7a1 100644
--- a/webrtc/modules/audio_processing/aec/aec_rdft_mips.cc
+++ b/webrtc/modules/audio_processing/utility/ooura_fft_mips.cc
@@ -8,10 +8,15 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
+#include "webrtc/modules/audio_processing/utility/ooura_fft.h"
+
+#include "webrtc/modules/audio_processing/utility/ooura_fft_tables_common.h"
 #include "webrtc/typedefs.h"
 
-static void bitrv2_128_mips(float* a) {
+namespace webrtc {
+
+#if defined(MIPS_FPU_LE)
+void bitrv2_128_mips(float* a) {
   // n is 128
   float xr, xi, yr, yi;
 
@@ -268,7 +273,7 @@ static void bitrv2_128_mips(float* a) {
   a[119] = xi;
 }
 
-static void cft1st_128_mips(float* a) {
+void cft1st_128_mips(float* a) {
   float f0, f1, f2, f3, f4, f5, f6, f7, f8, f9, f10, f11, f12, f13, f14;
   int a_ptr, p1_rdft, p2_rdft, count;
   const float* first = rdft_wk3ri_first;
@@ -517,7 +522,7 @@ static void cft1st_128_mips(float* a) {
   );
 }
 
-static void cftmdl_128_mips(float* a) {
+void cftmdl_128_mips(float* a) {
   float f0, f1, f2, f3, f4, f5, f6, f7, f8, f9, f10, f11, f12, f13, f14;
   int tmp_a, count;
   __asm __volatile (
@@ -803,12 +808,12 @@ static void cftmdl_128_mips(float* a) {
   );
 }
 
-static void cftfsub_128_mips(float* a) {
+void cftfsub_128_mips(float* a) {
   float f0, f1, f2, f3, f4, f5, f6, f7, f8;
   int tmp_a, count;
 
-  cft1st_128(a);
-  cftmdl_128(a);
+  cft1st_128_mips(a);
+  cftmdl_128_mips(a);
 
   __asm __volatile (
     ".set       push                                      \n\t"
@@ -861,12 +866,12 @@ static void cftfsub_128_mips(float* a) {
   );
 }
 
-static void cftbsub_128_mips(float* a) {
+void cftbsub_128_mips(float* a) {
   float f0, f1, f2, f3, f4, f5, f6, f7, f8;
   int tmp_a, count;
 
-  cft1st_128(a);
-  cftmdl_128(a);
+  cft1st_128_mips(a);
+  cftmdl_128_mips(a);
 
   __asm __volatile (
     ".set       push                                        \n\t"
@@ -919,7 +924,7 @@ static void cftbsub_128_mips(float* a) {
   );
 }
 
-static void rftfsub_128_mips(float* a) {
+void rftfsub_128_mips(float* a) {
   const float* c = rdft_w + 32;
   const float f0 = 0.5f;
   float* a1 = &a[2];
@@ -1046,7 +1051,7 @@ static void rftfsub_128_mips(float* a) {
   );
 }
 
-static void rftbsub_128_mips(float* a) {
+void rftbsub_128_mips(float* a) {
   const float *c = rdft_w + 32;
   const float f0 = 0.5f;
   float* a1 = &a[2];
@@ -1175,13 +1180,6 @@ static void rftbsub_128_mips(float* a) {
     : "memory"
   );
 }
+#endif
 
-void aec_rdft_init_mips(void) {
-  cft1st_128 = cft1st_128_mips;
-  cftmdl_128 = cftmdl_128_mips;
-  rftfsub_128 = rftfsub_128_mips;
-  rftbsub_128 = rftbsub_128_mips;
-  cftfsub_128 = cftfsub_128_mips;
-  cftbsub_128 = cftbsub_128_mips;
-  bitrv2_128 = bitrv2_128_mips;
-}
+}  // namespace webrtc
diff --git a/webrtc/modules/audio_processing/aec/aec_rdft_neon.cc b/webrtc/modules/audio_processing/utility/ooura_fft_neon.cc
similarity index 94%
rename from webrtc/modules/audio_processing/aec/aec_rdft_neon.cc
rename to webrtc/modules/audio_processing/utility/ooura_fft_neon.cc
index 43b6a68cd7..4ed043d518 100644
--- a/webrtc/modules/audio_processing/aec/aec_rdft_neon.cc
+++ b/webrtc/modules/audio_processing/utility/ooura_fft_neon.cc
@@ -14,15 +14,17 @@
  * Based on the sse2 version.
  */
 
-
-#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
+#include "webrtc/modules/audio_processing/utility/ooura_fft.h"
 
 #include <arm_neon.h>
 
-static const ALIGN16_BEG float ALIGN16_END
-    k_swap_sign[4] = {-1.f, 1.f, -1.f, 1.f};
+#include "webrtc/modules/audio_processing/utility/ooura_fft_tables_common.h"
+#include "webrtc/modules/audio_processing/utility/ooura_fft_tables_neon_sse2.h"
 
-static void cft1st_128_neon(float* a) {
+namespace webrtc {
+
+#if defined(WEBRTC_HAS_NEON)
+void cft1st_128_neon(float* a) {
   const float32x4_t vec_swap_sign = vld1q_f32((float32_t*)k_swap_sign);
   int j, k2;
 
@@ -71,7 +73,7 @@ static void cft1st_128_neon(float* a) {
   }
 }
 
-static void cftmdl_128_neon(float* a) {
+void cftmdl_128_neon(float* a) {
   int j;
   const int l = 8;
   const float32x4_t vec_swap_sign = vld1q_f32((float32_t*)k_swap_sign);
@@ -185,7 +187,7 @@ __inline static float32x4_t reverse_order_f32x4(float32x4_t in) {
   return vrev64q_f32(rev);
 }
 
-static void rftfsub_128_neon(float* a) {
+void rftfsub_128_neon(float* a) {
   const float* c = rdft_w + 32;
   int j1, j2;
   const float32x4_t mm_half = vdupq_n_f32(0.5f);
@@ -264,7 +266,7 @@ static void rftfsub_128_neon(float* a) {
   }
 }
 
-static void rftbsub_128_neon(float* a) {
+void rftbsub_128_neon(float* a) {
   const float* c = rdft_w + 32;
   int j1, j2;
   const float32x4_t mm_half = vdupq_n_f32(0.5f);
@@ -274,11 +276,11 @@ static void rftbsub_128_neon(float* a) {
   //    Note: commented number are indexes for the first iteration of the loop.
   for (j1 = 1, j2 = 2; j2 + 7 < 64; j1 += 4, j2 += 8) {
     // Load 'wk'.
-    const float32x4_t c_j1 = vld1q_f32(&c[j1]);         //  1,  2,  3,  4,
-    const float32x4_t c_k1 = vld1q_f32(&c[29 - j1]);    // 28, 29, 30, 31,
-    const float32x4_t wkrt = vsubq_f32(mm_half, c_k1);  // 28, 29, 30, 31,
-    const float32x4_t wkr_ = reverse_order_f32x4(wkrt); // 31, 30, 29, 28,
-    const float32x4_t wki_ = c_j1;                      //  1,  2,  3,  4,
+    const float32x4_t c_j1 = vld1q_f32(&c[j1]);          //  1,  2,  3,  4,
+    const float32x4_t c_k1 = vld1q_f32(&c[29 - j1]);     // 28, 29, 30, 31,
+    const float32x4_t wkrt = vsubq_f32(mm_half, c_k1);   // 28, 29, 30, 31,
+    const float32x4_t wkr_ = reverse_order_f32x4(wkrt);  // 31, 30, 29, 28,
+    const float32x4_t wki_ = c_j1;                       //  1,  2,  3,  4,
     // Load and shuffle 'a'.
     //   2,   4,   6,   8,   3,   5,   7,   9
     float32x4x2_t a_j2_p = vld2q_f32(&a[0 + j2]);
@@ -345,11 +347,6 @@ static void rftbsub_128_neon(float* a) {
   }
   a[65] = -a[65];
 }
+#endif
 
-void aec_rdft_init_neon(void) {
-  cft1st_128 = cft1st_128_neon;
-  cftmdl_128 = cftmdl_128_neon;
-  rftfsub_128 = rftfsub_128_neon;
-  rftbsub_128 = rftbsub_128_neon;
-}
-
+}  // namespace webrtc
diff --git a/webrtc/modules/audio_processing/aec/aec_rdft_sse2.cc b/webrtc/modules/audio_processing/utility/ooura_fft_sse2.cc
similarity index 85%
rename from webrtc/modules/audio_processing/aec/aec_rdft_sse2.cc
rename to webrtc/modules/audio_processing/utility/ooura_fft_sse2.cc
index b4e453ff53..03f6b31f0f 100644
--- a/webrtc/modules/audio_processing/aec/aec_rdft_sse2.cc
+++ b/webrtc/modules/audio_processing/utility/ooura_fft_sse2.cc
@@ -8,14 +8,32 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
+#include "webrtc/modules/audio_processing//utility/ooura_fft.h"
 
 #include <emmintrin.h>
 
-static const ALIGN16_BEG float ALIGN16_END
-    k_swap_sign[4] = {-1.f, 1.f, -1.f, 1.f};
+#include "webrtc/modules/audio_processing/utility/ooura_fft_tables_common.h"
+#include "webrtc/modules/audio_processing/utility/ooura_fft_tables_neon_sse2.h"
 
-static void cft1st_128_SSE2(float* a) {
+namespace webrtc {
+
+#if defined(WEBRTC_ARCH_X86_FAMILY)
+
+namespace {
+// These intrinsics were unavailable before VS 2008.
+// TODO(andrew): move to a common file.
+#if defined(_MSC_VER) && _MSC_VER < 1500
+static __inline __m128 _mm_castsi128_ps(__m128i a) {
+  return *(__m128*)&a;
+}
+static __inline __m128i _mm_castps_si128(__m128 a) {
+  return *(__m128i*)&a;
+}
+#endif
+
+}  // namespace
+
+void cft1st_128_SSE2(float* a) {
   const __m128 mm_swap_sign = _mm_load_ps(k_swap_sign);
   int j, k2;
 
@@ -78,7 +96,7 @@ static void cft1st_128_SSE2(float* a) {
   }
 }
 
-static void cftmdl_128_SSE2(float* a) {
+void cftmdl_128_SSE2(float* a) {
   const int l = 8;
   const __m128 mm_swap_sign = _mm_load_ps(k_swap_sign);
   int j0;
@@ -89,12 +107,12 @@ static void cftmdl_128_SSE2(float* a) {
     const __m128i a_08 = _mm_loadl_epi64((__m128i*)&a[j0 + 8]);
     const __m128i a_32 = _mm_loadl_epi64((__m128i*)&a[j0 + 32]);
     const __m128i a_40 = _mm_loadl_epi64((__m128i*)&a[j0 + 40]);
-    const __m128 a_00_32 = _mm_shuffle_ps(_mm_castsi128_ps(a_00),
-                                          _mm_castsi128_ps(a_32),
-                                          _MM_SHUFFLE(1, 0, 1, 0));
-    const __m128 a_08_40 = _mm_shuffle_ps(_mm_castsi128_ps(a_08),
-                                          _mm_castsi128_ps(a_40),
-                                          _MM_SHUFFLE(1, 0, 1, 0));
+    const __m128 a_00_32 =
+        _mm_shuffle_ps(_mm_castsi128_ps(a_00), _mm_castsi128_ps(a_32),
+                       _MM_SHUFFLE(1, 0, 1, 0));
+    const __m128 a_08_40 =
+        _mm_shuffle_ps(_mm_castsi128_ps(a_08), _mm_castsi128_ps(a_40),
+                       _MM_SHUFFLE(1, 0, 1, 0));
     __m128 x0r0_0i0_0r1_x0i1 = _mm_add_ps(a_00_32, a_08_40);
     const __m128 x1r0_1i0_1r1_x1i1 = _mm_sub_ps(a_00_32, a_08_40);
 
@@ -102,12 +120,12 @@ static void cftmdl_128_SSE2(float* a) {
     const __m128i a_24 = _mm_loadl_epi64((__m128i*)&a[j0 + 24]);
     const __m128i a_48 = _mm_loadl_epi64((__m128i*)&a[j0 + 48]);
     const __m128i a_56 = _mm_loadl_epi64((__m128i*)&a[j0 + 56]);
-    const __m128 a_16_48 = _mm_shuffle_ps(_mm_castsi128_ps(a_16),
-                                          _mm_castsi128_ps(a_48),
-                                          _MM_SHUFFLE(1, 0, 1, 0));
-    const __m128 a_24_56 = _mm_shuffle_ps(_mm_castsi128_ps(a_24),
-                                          _mm_castsi128_ps(a_56),
-                                          _MM_SHUFFLE(1, 0, 1, 0));
+    const __m128 a_16_48 =
+        _mm_shuffle_ps(_mm_castsi128_ps(a_16), _mm_castsi128_ps(a_48),
+                       _MM_SHUFFLE(1, 0, 1, 0));
+    const __m128 a_24_56 =
+        _mm_shuffle_ps(_mm_castsi128_ps(a_24), _mm_castsi128_ps(a_56),
+                       _MM_SHUFFLE(1, 0, 1, 0));
     const __m128 x2r0_2i0_2r1_x2i1 = _mm_add_ps(a_16_48, a_24_56);
     const __m128 x3r0_3i0_3r1_x3i1 = _mm_sub_ps(a_16_48, a_24_56);
 
@@ -163,12 +181,12 @@ static void cftmdl_128_SSE2(float* a) {
       const __m128i a_08 = _mm_loadl_epi64((__m128i*)&a[j0 + 8]);
       const __m128i a_32 = _mm_loadl_epi64((__m128i*)&a[j0 + 32]);
       const __m128i a_40 = _mm_loadl_epi64((__m128i*)&a[j0 + 40]);
-      const __m128 a_00_32 = _mm_shuffle_ps(_mm_castsi128_ps(a_00),
-                                            _mm_castsi128_ps(a_32),
-                                            _MM_SHUFFLE(1, 0, 1, 0));
-      const __m128 a_08_40 = _mm_shuffle_ps(_mm_castsi128_ps(a_08),
-                                            _mm_castsi128_ps(a_40),
-                                            _MM_SHUFFLE(1, 0, 1, 0));
+      const __m128 a_00_32 =
+          _mm_shuffle_ps(_mm_castsi128_ps(a_00), _mm_castsi128_ps(a_32),
+                         _MM_SHUFFLE(1, 0, 1, 0));
+      const __m128 a_08_40 =
+          _mm_shuffle_ps(_mm_castsi128_ps(a_08), _mm_castsi128_ps(a_40),
+                         _MM_SHUFFLE(1, 0, 1, 0));
       __m128 x0r0_0i0_0r1_x0i1 = _mm_add_ps(a_00_32, a_08_40);
       const __m128 x1r0_1i0_1r1_x1i1 = _mm_sub_ps(a_00_32, a_08_40);
 
@@ -176,22 +194,21 @@ static void cftmdl_128_SSE2(float* a) {
       const __m128i a_24 = _mm_loadl_epi64((__m128i*)&a[j0 + 24]);
       const __m128i a_48 = _mm_loadl_epi64((__m128i*)&a[j0 + 48]);
       const __m128i a_56 = _mm_loadl_epi64((__m128i*)&a[j0 + 56]);
-      const __m128 a_16_48 = _mm_shuffle_ps(_mm_castsi128_ps(a_16),
-                                            _mm_castsi128_ps(a_48),
-                                            _MM_SHUFFLE(1, 0, 1, 0));
-      const __m128 a_24_56 = _mm_shuffle_ps(_mm_castsi128_ps(a_24),
-                                            _mm_castsi128_ps(a_56),
-                                            _MM_SHUFFLE(1, 0, 1, 0));
+      const __m128 a_16_48 =
+          _mm_shuffle_ps(_mm_castsi128_ps(a_16), _mm_castsi128_ps(a_48),
+                         _MM_SHUFFLE(1, 0, 1, 0));
+      const __m128 a_24_56 =
+          _mm_shuffle_ps(_mm_castsi128_ps(a_24), _mm_castsi128_ps(a_56),
+                         _MM_SHUFFLE(1, 0, 1, 0));
       const __m128 x2r0_2i0_2r1_x2i1 = _mm_add_ps(a_16_48, a_24_56);
       const __m128 x3r0_3i0_3r1_x3i1 = _mm_sub_ps(a_16_48, a_24_56);
 
       const __m128 xx = _mm_add_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
       const __m128 xx1 = _mm_sub_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
       const __m128 xx2 = _mm_mul_ps(xx1, wk2rv);
-      const __m128 xx3 =
-          _mm_mul_ps(wk2iv,
-                     _mm_castsi128_ps(_mm_shuffle_epi32(
-                         _mm_castps_si128(xx1), _MM_SHUFFLE(2, 3, 0, 1))));
+      const __m128 xx3 = _mm_mul_ps(
+          wk2iv, _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(xx1),
+                                                    _MM_SHUFFLE(2, 3, 0, 1))));
       const __m128 xx4 = _mm_add_ps(xx2, xx3);
 
       const __m128 x3i0_3r0_3i1_x3r1 = _mm_castsi128_ps(_mm_shuffle_epi32(
@@ -202,16 +219,14 @@ static void cftmdl_128_SSE2(float* a) {
 
       const __m128 xx10 = _mm_mul_ps(x1_x3_add, wk1rv);
       const __m128 xx11 = _mm_mul_ps(
-          wk1iv,
-          _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(x1_x3_add),
-                                             _MM_SHUFFLE(2, 3, 0, 1))));
+          wk1iv, _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(x1_x3_add),
+                                                    _MM_SHUFFLE(2, 3, 0, 1))));
       const __m128 xx12 = _mm_add_ps(xx10, xx11);
 
       const __m128 xx20 = _mm_mul_ps(x1_x3_sub, wk3rv);
       const __m128 xx21 = _mm_mul_ps(
-          wk3iv,
-          _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(x1_x3_sub),
-                                             _MM_SHUFFLE(2, 3, 0, 1))));
+          wk3iv, _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(x1_x3_sub),
+                                                    _MM_SHUFFLE(2, 3, 0, 1))));
       const __m128 xx22 = _mm_add_ps(xx20, xx21);
 
       _mm_storel_epi64((__m128i*)&a[j0 + 0], _mm_castps_si128(xx));
@@ -237,13 +252,13 @@ static void cftmdl_128_SSE2(float* a) {
   }
 }
 
-static void rftfsub_128_SSE2(float* a) {
+void rftfsub_128_SSE2(float* a) {
   const float* c = rdft_w + 32;
   int j1, j2, k1, k2;
   float wkr, wki, xr, xi, yr, yi;
 
-  static const ALIGN16_BEG float ALIGN16_END
-      k_half[4] = {0.5f, 0.5f, 0.5f, 0.5f};
+  static const ALIGN16_BEG float ALIGN16_END k_half[4] = {0.5f, 0.5f, 0.5f,
+                                                          0.5f};
   const __m128 mm_half = _mm_load_ps(k_half);
 
   // Vectorized code (four at once).
@@ -327,13 +342,13 @@ static void rftfsub_128_SSE2(float* a) {
   }
 }
 
-static void rftbsub_128_SSE2(float* a) {
+void rftbsub_128_SSE2(float* a) {
   const float* c = rdft_w + 32;
   int j1, j2, k1, k2;
   float wkr, wki, xr, xi, yr, yi;
 
-  static const ALIGN16_BEG float ALIGN16_END
-      k_half[4] = {0.5f, 0.5f, 0.5f, 0.5f};
+  static const ALIGN16_BEG float ALIGN16_END k_half[4] = {0.5f, 0.5f, 0.5f,
+                                                          0.5f};
   const __m128 mm_half = _mm_load_ps(k_half);
 
   a[1] = -a[1];
@@ -418,10 +433,6 @@ static void rftbsub_128_SSE2(float* a) {
   }
   a[65] = -a[65];
 }
+#endif
 
-void aec_rdft_init_sse2(void) {
-  cft1st_128 = cft1st_128_SSE2;
-  cftmdl_128 = cftmdl_128_SSE2;
-  rftfsub_128 = rftfsub_128_SSE2;
-  rftbsub_128 = rftbsub_128_SSE2;
-}
+}  // namespace webrtc
diff --git a/webrtc/modules/audio_processing/utility/ooura_fft_tables_common.h b/webrtc/modules/audio_processing/utility/ooura_fft_tables_common.h
new file mode 100644
index 0000000000..548027cf22
--- /dev/null
+++ b/webrtc/modules/audio_processing/utility/ooura_fft_tables_common.h
@@ -0,0 +1,54 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_UTILITY_OOURA_FFT_TABLES_COMMON_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_UTILITY_OOURA_FFT_TABLES_COMMON_H_
+
+#include "webrtc/modules/audio_processing/utility/ooura_fft.h"
+
+namespace webrtc {
+
+// This tables used to be computed at run-time. For example, refer to:
+// https://code.google.com/p/webrtc/source/browse/trunk/webrtc/modules/audio_processing/utility/apm_rdft.c?r=6564
+// to see the initialization code.
+// Constants shared by all paths (C, SSE2, NEON).
+const float rdft_w[64] = {
+    1.0000000000f, 0.0000000000f, 0.7071067691f, 0.7071067691f, 0.9238795638f,
+    0.3826834559f, 0.3826834559f, 0.9238795638f, 0.9807852507f, 0.1950903237f,
+    0.5555702448f, 0.8314695954f, 0.8314695954f, 0.5555702448f, 0.1950903237f,
+    0.9807852507f, 0.9951847196f, 0.0980171412f, 0.6343933344f, 0.7730104327f,
+    0.8819212914f, 0.4713967443f, 0.2902846634f, 0.9569403529f, 0.9569403529f,
+    0.2902846634f, 0.4713967443f, 0.8819212914f, 0.7730104327f, 0.6343933344f,
+    0.0980171412f, 0.9951847196f, 0.7071067691f, 0.4993977249f, 0.4975923598f,
+    0.4945882559f, 0.4903926253f, 0.4850156307f, 0.4784701765f, 0.4707720280f,
+    0.4619397819f, 0.4519946277f, 0.4409606457f, 0.4288643003f, 0.4157347977f,
+    0.4016037583f, 0.3865052164f, 0.3704755902f, 0.3535533845f, 0.3357794881f,
+    0.3171966672f, 0.2978496552f, 0.2777851224f, 0.2570513785f, 0.2356983721f,
+    0.2137775421f, 0.1913417280f, 0.1684449315f, 0.1451423317f, 0.1214900985f,
+    0.0975451618f, 0.0733652338f, 0.0490085706f, 0.0245338380f,
+};
+
+// Constants used by the C and MIPS paths.
+const float rdft_wk3ri_first[16] = {
+    1.000000000f, 0.000000000f, 0.382683456f,  0.923879564f,
+    0.831469536f, 0.555570245f, -0.195090353f, 0.980785251f,
+    0.956940353f, 0.290284693f, 0.098017156f,  0.995184720f,
+    0.634393334f, 0.773010492f, -0.471396863f, 0.881921172f,
+};
+const float rdft_wk3ri_second[16] = {
+    -0.707106769f, 0.707106769f,  -0.923879564f, -0.382683456f,
+    -0.980785251f, 0.195090353f,  -0.555570245f, -0.831469536f,
+    -0.881921172f, 0.471396863f,  -0.773010492f, -0.634393334f,
+    -0.995184720f, -0.098017156f, -0.290284693f, -0.956940353f,
+};
+
+}  // namespace webrtc
+
+#endif  // WEBRTC_MODULES_AUDIO_PROCESSING_UTILITY_OOURA_FFT_TABLES_COMMON_H_
diff --git a/webrtc/modules/audio_processing/utility/ooura_fft_tables_neon_sse2.h b/webrtc/modules/audio_processing/utility/ooura_fft_tables_neon_sse2.h
new file mode 100644
index 0000000000..1ed646d6f6
--- /dev/null
+++ b/webrtc/modules/audio_processing/utility/ooura_fft_tables_neon_sse2.h
@@ -0,0 +1,94 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_UTILITY_OOURA_FFT_TABLES_NEON_SSE2_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_UTILITY_OOURA_FFT_TABLES_NEON_SSE2_H_
+
+#include "webrtc/modules/audio_processing/utility/ooura_fft.h"
+
+#ifdef _MSC_VER /* visual c++ */
+#define ALIGN16_BEG __declspec(align(16))
+#define ALIGN16_END
+#else /* gcc or icc */
+#define ALIGN16_BEG
+#define ALIGN16_END __attribute__((aligned(16)))
+#endif
+
+namespace webrtc {
+
+// These tables used to be computed at run-time. For example, refer to:
+// https://code.google.com/p/webrtc/source/browse/trunk/webrtc/modules/audio_processing/utility/apm_rdft.c?r=6564
+// to see the initialization code.
+#if defined(WEBRTC_ARCH_X86_FAMILY) || defined(WEBRTC_HAS_NEON)
+// Constants used by SSE2 and NEON but initialized in the C path.
+const ALIGN16_BEG float ALIGN16_END k_swap_sign[4] = {-1.f, 1.f, -1.f, 1.f};
+
+ALIGN16_BEG const float ALIGN16_END rdft_wk1r[32] = {
+    1.000000000f, 1.000000000f, 0.707106769f, 0.707106769f, 0.923879564f,
+    0.923879564f, 0.382683456f, 0.382683456f, 0.980785251f, 0.980785251f,
+    0.555570245f, 0.555570245f, 0.831469595f, 0.831469595f, 0.195090324f,
+    0.195090324f, 0.995184720f, 0.995184720f, 0.634393334f, 0.634393334f,
+    0.881921291f, 0.881921291f, 0.290284663f, 0.290284663f, 0.956940353f,
+    0.956940353f, 0.471396744f, 0.471396744f, 0.773010433f, 0.773010433f,
+    0.098017141f, 0.098017141f,
+};
+ALIGN16_BEG const float ALIGN16_END rdft_wk2r[32] = {
+    1.000000000f,  1.000000000f,  -0.000000000f, -0.000000000f, 0.707106769f,
+    0.707106769f,  -0.707106769f, -0.707106769f, 0.923879564f,  0.923879564f,
+    -0.382683456f, -0.382683456f, 0.382683456f,  0.382683456f,  -0.923879564f,
+    -0.923879564f, 0.980785251f,  0.980785251f,  -0.195090324f, -0.195090324f,
+    0.555570245f,  0.555570245f,  -0.831469595f, -0.831469595f, 0.831469595f,
+    0.831469595f,  -0.555570245f, -0.555570245f, 0.195090324f,  0.195090324f,
+    -0.980785251f, -0.980785251f,
+};
+ALIGN16_BEG const float ALIGN16_END rdft_wk3r[32] = {
+    1.000000000f,  1.000000000f,  -0.707106769f, -0.707106769f, 0.382683456f,
+    0.382683456f,  -0.923879564f, -0.923879564f, 0.831469536f,  0.831469536f,
+    -0.980785251f, -0.980785251f, -0.195090353f, -0.195090353f, -0.555570245f,
+    -0.555570245f, 0.956940353f,  0.956940353f,  -0.881921172f, -0.881921172f,
+    0.098017156f,  0.098017156f,  -0.773010492f, -0.773010492f, 0.634393334f,
+    0.634393334f,  -0.995184720f, -0.995184720f, -0.471396863f, -0.471396863f,
+    -0.290284693f, -0.290284693f,
+};
+ALIGN16_BEG const float ALIGN16_END rdft_wk1i[32] = {
+    -0.000000000f, 0.000000000f,  -0.707106769f, 0.707106769f,  -0.382683456f,
+    0.382683456f,  -0.923879564f, 0.923879564f,  -0.195090324f, 0.195090324f,
+    -0.831469595f, 0.831469595f,  -0.555570245f, 0.555570245f,  -0.980785251f,
+    0.980785251f,  -0.098017141f, 0.098017141f,  -0.773010433f, 0.773010433f,
+    -0.471396744f, 0.471396744f,  -0.956940353f, 0.956940353f,  -0.290284663f,
+    0.290284663f,  -0.881921291f, 0.881921291f,  -0.634393334f, 0.634393334f,
+    -0.995184720f, 0.995184720f,
+};
+ALIGN16_BEG const float ALIGN16_END rdft_wk2i[32] = {
+    -0.000000000f, 0.000000000f,  -1.000000000f, 1.000000000f,  -0.707106769f,
+    0.707106769f,  -0.707106769f, 0.707106769f,  -0.382683456f, 0.382683456f,
+    -0.923879564f, 0.923879564f,  -0.923879564f, 0.923879564f,  -0.382683456f,
+    0.382683456f,  -0.195090324f, 0.195090324f,  -0.980785251f, 0.980785251f,
+    -0.831469595f, 0.831469595f,  -0.555570245f, 0.555570245f,  -0.555570245f,
+    0.555570245f,  -0.831469595f, 0.831469595f,  -0.980785251f, 0.980785251f,
+    -0.195090324f, 0.195090324f,
+};
+ALIGN16_BEG const float ALIGN16_END rdft_wk3i[32] = {
+    -0.000000000f, 0.000000000f,  -0.707106769f, 0.707106769f,  -0.923879564f,
+    0.923879564f,  0.382683456f,  -0.382683456f, -0.555570245f, 0.555570245f,
+    -0.195090353f, 0.195090353f,  -0.980785251f, 0.980785251f,  0.831469536f,
+    -0.831469536f, -0.290284693f, 0.290284693f,  -0.471396863f, 0.471396863f,
+    -0.995184720f, 0.995184720f,  0.634393334f,  -0.634393334f, -0.773010492f,
+    0.773010492f,  0.098017156f,  -0.098017156f, -0.881921172f, 0.881921172f,
+    0.956940353f,  -0.956940353f,
+};
+ALIGN16_BEG const float ALIGN16_END cftmdl_wk1r[4] = {
+    0.707106769f, 0.707106769f, 0.707106769f, -0.707106769f,
+};
+#endif
+
+}  // namespace webrtc
+
+#endif  // WEBRTC_MODULES_AUDIO_PROCESSING_UTILITY_OOURA_FFT_TABLES_NEON_SSE2_H_