|
Chris@3
|
1 <html>
|
|
Chris@7
|
2 <head>
|
|
Chris@7
|
3
|
|
Chris@3
|
4 <meta charset="UTF-8">
|
|
Chris@1
|
5
|
|
Chris@3
|
6 <style type="text/css">
|
|
Chris@3
|
7 body { margin: 5%; }
|
|
Chris@3
|
8 table, td, th { border: 0.1em solid #e0e0e0; border-collapse: collapse }
|
|
Chris@3
|
9 td, th { padding: 0.5em }
|
|
Chris@3
|
10 </style>
|
|
Chris@3
|
11
|
|
Chris@3
|
12 <script src="nayuki/fft.js"></script>
|
|
Chris@17
|
13 <script src="nayuki-obj/fft.js"></script>
|
|
Chris@3
|
14 <script src="fft.js/lib/complex.js"></script>
|
|
Chris@3
|
15 <script src="jsfft/lib/complex_array.js"></script>
|
|
Chris@3
|
16 <script src="jsfft/lib/fft.js"></script>
|
|
Chris@7
|
17 <script src="cross/Cross.js"></script>
|
|
Chris@7
|
18 <script src="cross/FFT.js"></script>
|
|
Chris@8
|
19 <script src="kissfft/KissFFT.js"></script>
|
|
Chris@8
|
20 <script src="kissfft/FFT.js"></script>
|
|
Chris@3
|
21 <script src="test.js"></script>
|
|
Chris@1
|
22
|
|
Chris@7
|
23 </head>
|
|
Chris@3
|
24 <body>
|
|
Chris@2
|
25
|
|
Chris@3
|
26 <h3>Results</h3>
|
|
Chris@1
|
27
|
|
Chris@3
|
28 <p id="test-description"></p>
|
|
Chris@3
|
29
|
|
Chris@3
|
30 <table>
|
|
Chris@3
|
31 <tr>
|
|
Chris@3
|
32 <th>Implementation</th><th>Result</th><th>Time (first half)</th><th>Time (second half)</th><th>Rate (second half)</th>
|
|
Chris@3
|
33 </tr>
|
|
Chris@3
|
34 <tr>
|
|
Chris@3
|
35 <td>Nayuki</td><td id="nayuki-result"></td><td id="nayuki-1"></td><td id="nayuki-2"></td><td id="nayuki-itr"></td>
|
|
Chris@17
|
36 </tr><tr>
|
|
Chris@17
|
37 <td>Nayuki (obj)</td><td id="nayukiobj-result"></td><td id="nayukiobj-1"></td><td id="nayukiobj-2"></td><td id="nayukiobj-itr"></td>
|
|
Chris@17
|
38 </tr><tr>
|
|
Chris@3
|
39 <td>Nockert</td><td id="nockert-result"></td><td id="nockert-1"></td><td id="nockert-2"></td><td id="nockert-itr"></td>
|
|
Chris@3
|
40 </tr><tr>
|
|
Chris@3
|
41 <td>Dntj</td><td id="dntj-result"></td><td id="dntj-1"></td><td id="dntj-2"></td><td id="dntj-itr"></td>
|
|
Chris@7
|
42 </tr><tr>
|
|
Chris@7
|
43 <td>Cross</td><td id="cross-result"></td><td id="cross-1"></td><td id="cross-2"></td><td id="cross-itr"></td>
|
|
Chris@8
|
44 </tr><tr>
|
|
Chris@8
|
45 <td>KissFFT</td><td id="kissfft-result"></td><td id="kissfft-1"></td><td id="kissfft-2"></td><td id="kissfft-itr"></td>
|
|
Chris@3
|
46 </tr>
|
|
Chris@3
|
47 </table>
|
|
Chris@1
|
48
|
|
Chris@3
|
49 <h3>Notes</h3>
|
|
Chris@1
|
50
|
|
Chris@3
|
51 <ul>
|
|
Chris@3
|
52 <li><b>Nayuki</b>: in-place single-precision complex-complex</li>
|
|
Chris@17
|
53 <li><b>Nayuki (obj)</b>: Nayuki with the sin/cos tables pre-calculated on object construction</li>
|
|
Chris@3
|
54 <li><b>Nockert</b>: double-precision real-complex</li>
|
|
Chris@17
|
55 <li><b>Dntj</b>: double-precision complex-complex. Forward
|
|
Chris@10
|
56 transform is scaled and I've scaled it back again here, which may
|
|
Chris@10
|
57 introduce rounding error.</li>
|
|
Chris@10
|
58 <li><b>Cross</b>: double-precision real-complex in C, compiled
|
|
Chris@10
|
59 with Emscripten. This is considered a slow implementation amongst
|
|
Chris@10
|
60 native code ones.</li>
|
|
Chris@10
|
61 <li><b>KissFFT</b>: single-precision real-complex in C, compiled
|
|
Chris@10
|
62 with Emscripten. This should be faster than Cross. Despite its
|
|
Chris@10
|
63 name, it is the most sophisticated implementation here.</li>
|
|
Chris@3
|
64 </ul>
|
|
Chris@3
|
65
|
|
Chris@3
|
66 <h3>Rationale</h3>
|
|
Chris@1
|
67
|
|
Chris@3
|
68 <p>If 2150 iterations of real-to-complex FFT of size 2048 takes less
|
|
Chris@3
|
69 than 10 seconds, then we may be able to make a high quality
|
|
Chris@3
|
70 real-time phase vocoder (just).</p>
|
|
Chris@1
|
71
|
|
Chris@3
|
72 <p>A phase-vocoder of course must use overlapped windowed FFT
|
|
Chris@3
|
73 (although you can choose the size, within limits), IFFT, and
|
|
Chris@3
|
74 cartesian-polar conversion to calculate the phase for the
|
|
Chris@3
|
75 instantaneous frequency.</p>
|
|
Chris@1
|
76
|
|
Chris@3
|
77 <p>A reasonable estimate of CPU cost for the whole thing is
|
|
Chris@3
|
78 somewhere around 10x the cost of simple non-overlapping short-time
|
|
Chris@3
|
79 forward Fourier transforms across the signal. </p>
|
|
Chris@1
|
80
|
|
Chris@3
|
81 <p>2150 iterations corresponds to 100 seconds of audio
|
|
Chris@3
|
82 non-overlapped at 44.1kHz, so if that takes less than 10 seconds,
|
|
Chris@3
|
83 then in theory we might be OK.</p>
|
|
Chris@1
|
84
|
|
Chris@3
|
85 </body>
|
|
Chris@1
|
86
|
