Use expf() instead of an approximation
An approximation of exp function is used in Compress,
but the argument can exceed +-1.0 when the input
signal or the target gain is drastically changed,
then the output of the function can becomes inaccurate.
Use expf() to get accurate output since its
performance is enough based on the benchmark.
Bug: 314374972
Test: Change targetgain when using loudnessenhancer
Change-Id: I47795345655b4ad8becb0ee18ef981eb40ff847f
diff --git a/media/libeffects/loudness/dsp/core/dynamic_range_compression.cpp b/media/libeffects/loudness/dsp/core/dynamic_range_compression.cpp
index 578f58a..33f6779 100644
--- a/media/libeffects/loudness/dsp/core/dynamic_range_compression.cpp
+++ b/media/libeffects/loudness/dsp/core/dynamic_range_compression.cpp
@@ -89,8 +89,7 @@
} else {
state_ = alpha_release_ * state_ + (1.0f - alpha_release_) * cv;
}
- compressor_gain_ *=
- math::ExpApproximationViaTaylorExpansionOrder5(state_ - prev_state);
+ compressor_gain_ *= expf(state_ - prev_state);
x *= compressor_gain_;
if (x > kFixedPointLimit) {
return kFixedPointLimit;
@@ -118,8 +117,7 @@
} else {
state_ = alpha_release_ * state_ + (1.0f - alpha_release_) * cv;
}
- compressor_gain_ *=
- math::ExpApproximationViaTaylorExpansionOrder5(state_ - prev_state);
+ compressor_gain_ *= expf(state_ - prev_state);
*x1 *= compressor_gain_;
if (*x1 > kFixedPointLimit) {
*x1 = kFixedPointLimit;