The regular USoundWave isn’t well-suited for dynamically filling in audio data, but is more geared towards serialization on the editor side for later runtime use, so this runtime sound wave creation behavior can vary between different UE versions. From my experience, it’s common for its behavior to change across versions, including differences in how it handles compressed audio buffers and the ways for getting and populating audio data.
I recommend decoding the audio data to PCM manually and then queuing it into USoundWaveProcedural (which is inherited from USoundWave), as this approach maintains consistent behavior across different engine versions. For example, RuntimeAudioImporter uses the same approach and works consistently and correctly from UE 4.24 through the latest version, 5.4, and it can handle WAV formats. You can check it out here: https://www.fab.com/listings/66e0d72e-982f-4d9e-aaaf-13a1d22efad1