Here are the results, proofing that with the WASAPI architecture I got true audio 24 bit samples under Windows 7:
Architecture: MME (default for PyAudio)
Requested resolution: 16 bit

Architecture: WASAPI
Requested resolution: 16 bit
Architecture: WASAPI

Architecture: MME (default for PyAudio)
Requested resolution: 16 bit
Architecture: MME (default for PyAudio)
Requested resolution: 24 bit
Note that the returned data has still only 16 bit precission and is zero padded to 24 bit.

Architecture: WASAPI
Requested resolution: 16 bit
Architecture: WASAPI
Requested resolution: 24 bit
This is how it should be. Note the much better resolution. Now the analog part of the soundcard is determining the noise floor and not the quantization noise.

No comments:
Post a Comment