diff --git a/android/native/jni/Android.mk b/android/native/jni/Android.mk
index de472c0a25..b36ac2b57f 100644
--- a/android/native/jni/Android.mk
+++ b/android/native/jni/Android.mk
@@ -20,6 +20,7 @@ ifeq ($(TARGET_ARCH_ABI),armeabi-v7a)
 
 ifeq ($(HAVE_NEON),1)
 LOCAL_CFLAGS += -DHAVE_NEON
+LOCAL_SRC_FILES += ../../../audio/utils_neon.S.neon
 endif
 
 ifeq ($(HAVE_SINC),1)
diff --git a/audio/utils.c b/audio/utils.c
index 4fa8a3f89e..1781a4a06a 100644
--- a/audio/utils.c
+++ b/audio/utils.c
@@ -133,6 +133,30 @@ void audio_convert_float_to_s16_altivec(int16_t *out,
    else
       audio_convert_float_to_s16_C(out, in, samples);
 }
+#elif HAVE_NEON
+void audio_convert_s16_float_asm(float *out, const int16_t *in, size_t samples);
+void audio_convert_s16_to_float_neon(float *out, const int16_t *in, size_t samples,
+      float gain)
+{
+   (void)gain; // gain is ignored for now.
 
+   size_t aligned_samples = samples & 7;
+   audio_convert_s16_float_asm(out, in, aligned_samples);
+
+   // Could do all conversion in ASM, but keep it simple for now.
+   audio_convert_s16_to_float_C(out + aligned_samples, in + aligned_samples,
+         samples - aligned_samples);
+}
+
+void audio_convert_float_s16_asm(int16_t *out, const float *in, size_t samples);
+void audio_convert_float_to_s16_neon(int16_t *out, const float *in, size_t samples)
+{
+   (void)gain; // gain is ignored for now.
+
+   size_t aligned_samples = samples & 7;
+   audio_convert_float_s16_asm(out, in, aligned_samples);
+   audio_convert_float_to_s16_C(out + aligned_samples, in + aligned_samples,
+         samples - aligned_samples);
+}
 #endif
 
diff --git a/audio/utils.h b/audio/utils.h
index 26238881cc..ac7ca65caf 100644
--- a/audio/utils.h
+++ b/audio/utils.h
@@ -39,6 +39,16 @@ void audio_convert_s16_to_float_altivec(float *out,
 void audio_convert_float_to_s16_altivec(int16_t *out,
       const float *in, size_t samples);
 
+#elif defined(HAVE_NEON)
+#define audio_convert_s16_to_float audio_convert_s16_to_float_neon
+#define audio_convert_float_to_s16 audio_convert_float_to_s16_neon
+
+void audio_convert_s16_to_float_neon(float *out,
+      const int16_t *in, size_t samples, float gain);
+
+void audio_convert_float_to_s16_neon(int16_t *out,
+      const float *in, size_t samples);
+
 #else
 #define audio_convert_s16_to_float audio_convert_s16_to_float_C
 #define audio_convert_float_to_s16 audio_convert_float_to_s16_C
diff --git a/audio/utils_neon.S b/audio/utils_neon.S
new file mode 100644
index 0000000000..c4834a9249
--- /dev/null
+++ b/audio/utils_neon.S
@@ -0,0 +1,86 @@
+/*  RetroArch - A frontend for libretro.
+ *  Copyright (C) 2010-2012 - Hans-Kristian Arntzen
+ * 
+ *  RetroArch is free software: you can redistribute it and/or modify it under the terms
+ *  of the GNU General Public License as published by the Free Software Found-
+ *  ation, either version 3 of the License, or (at your option) any later version.
+ *
+ *  RetroArch is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
+ *  without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
+ *  PURPOSE.  See the GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License along with RetroArch.
+ *  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+.arm
+
+.align 4
+.global audio_convert_s16_float_asm
+# audio_convert_s16_float_asm(float *out, const int16_t *in, size_t samples)
+audio_convert_s16_float_asm:
+   # Hacky way to get a constant of 2^-15.
+   # Might be faster to just load a constant from memory.
+   # It's just done once however ...
+   vmov.f32 q5, #0.25
+   vmul.f32 q5, q5, q5
+   vmul.f32 q5, q5, q5
+   vmul.f32 q5, q5, q5
+   vadd.f32 q5, q5, q5
+
+1:
+   # Preload here?
+   vld1.s16 {q0}, [r1]!
+
+   # Widen to 32-bit
+   vmovl.s16 q1, d0
+   vmovl.s16 q2, d1
+
+   # Convert to float
+   vcvt.f32.s32 q3, q1
+   vcvt.f32.s32 q4, q2
+
+   vmul.f32 q3, q3, q5
+   vmul.f32 q4, q4, q5
+
+   vst1.f32 {q3-q4}, [r0]!
+
+   # Guaranteed to get samples in multiples of 8.
+   subs r2, r2, #8
+   bne 1b
+
+   bx lr
+
+.align 4
+.global audio_convert_float_s16_asm
+# audio_convert_float_s16_asm(int16_t *out, const float *in, size_t samples)
+audio_convert_float_s16_asm:
+   # Hacky way to get a constant of 2^15.
+   # ((2^4)^2)^2 * 0.5 = 2^15
+   vmov.f32 q5, #16.0
+   vmov.f32 q6, #0.5
+   vmul.f32 q5, q5, q5
+   vmul.f32 q5, q5, q5
+   vmul.f32 q5, q5, q6
+
+1:
+   # Preload here?
+   vld1.f32 {q0-q1}, [r1]!
+
+   vmul.f32 q0, q0, q5
+   vmul.f32 q1, q1, q5
+
+   vcvt.s32.f32 q0, q0
+   vcvt.s32.f32 q1, q1
+
+   vqmovn.s32 d4, q0
+   vqmovn.s32 d5, q1
+
+   vst1.f32 {d4-d5}, [r0]!
+
+   # Guaranteed to get samples in multiples of 8.
+   subs r2, r2, #8
+   bne 1b
+
+   bx lr
+