相信接觸過數字音頻的朋友, 都會有過同樣的疑問, 當然不會有絕對答案, 因為兩邊也會有支持者, 但我們最重要的是 - 結果 !! 那些資料可看成是見識、參考和一種大約的概念 (例如 浮點運算的好處是容易編寫, 定點較準確, 但每一步都要親力親為去寫), 因為當中可以牽涉了深層的物理學, 正如 bob L 所說, 一般編程人員也不容易理解這類音頻軟件的其中關係, 也超越了我們需要知道的範圍, 所以最好的方法是聽聽出來的結果。
如果軟件內部是 32BIT 浮點計算, 理論上無論是 16或 24BIT的文件, 也是以內部32BIT浮點這方式去處理。
以下是從SAW FORUM轉過來的, 也可了解一下音頻軟件的數據流程和內部處理運算選擇的不同:-
* the bit depth of the internal processing math (which can either be floating point or fixed point). What this means is how many digits calculations occur at - bigger bit depths (i.e. more digits to work with) means more precise calculations. SAW uses 64bit fixed point math for its multiplication and division. Most other native PC based DAW apps use 32bit floating point for this. The SAWStudio built in eq, the Anwida and JMS Audioware SAW native plugins, and some of the filter codes used in the Sonoris SAW native plugins use 64bit floating point math.
* the bit depth of the returned figures after a DSP has been performed - i.e. after a signal process is done, how many digits are passed to the next signal process prior to output. In SAW this is at 32bit fixed point using the full DWORD. With most other PC based DAW appps this is a 24bit + 8bit mantissa 32bit IEEE floating point file.
[ 本帖最后由 himhui 于 07-11-1 22:20 编辑 ]