-ffast-math changes the floating point rounding mode globally, which
is undesirable for shared libraries. 

see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55522
for details.

--- soundtouch/CMakeLists.txt	2021-09-07 17:26:53.000000000 +0200
+++ soundtouch/CMakeLists.txt	2022-09-11 10:52:23.945515340 +0200
@@ -3,13 +3,6 @@
 
 include(GNUInstallDirs)
 
-if(MSVC)
-  set(COMPILE_DEFINITIONS /O2 /fp:fast)
-  set(COMPILE_OPTIONS )
-else()
-  set(COMPILE_OPTIONS -Ofast)
-endif()
-
 #####################
 # SoundTouch library
 
--- soundtouch/configure.ac	2021-09-07 17:26:53.000000000 +0200
+++ soundtouch/configure.ac	2022-09-11 10:52:44.845868245 +0200
@@ -33,7 +33,7 @@
 
 # Compiler flags. Apply -ffast-math to allow gcc autovectorization
 # generate effective SIMD code.
-CXXFLAGS+=" -O3 -ffast-math"
+CXXFLAGS+=" -O3 "
 
 # Set AR_FLAGS to avoid build warning "ar: `u' modifier ignored since `D' is the default (see `U')"
 AR_FLAGS='cr'