Applying cosine window before and after spectral processing

Author: sadko4u

src/main/misc/windows.cpp

@@ -251,7 +251,7 @@ namespace lsp
                 if (n == 0)
                     return;
 
-                float f         = M_PI / n;
+                float f         = M_PI / (n - 1);
                 for (size_t i=0; i<n; ++i)
                 {
                     float a     = sinf(f * i);

src/main/util/SpectralProcessor.cpp

@@ -121,7 +121,7 @@ namespace lsp
             pFftBuf         = &pInBuf[buf_size];
 
             // Clear buffers and reset pointers
-            windows::sqr_cosine(pWnd, buf_size);
+            windows::cosine(pWnd, buf_size);
             dsp::fill_zero(pOutBuf, buf_size*4);     // OutBuf + InBuf + Fft(x2)
             nOffset         = buf_size * (fPhase * 0.5f);
 
@@ -161,19 +161,20 @@ namespace lsp
                     if (pFunc != NULL)
                     {
                         // Perform FFT and processing
-                        dsp::pcomplex_r2c(pFftBuf, pInBuf, buf_size);       // Convert from real to packed complex
-                        dsp::packed_direct_fft(pFftBuf, pFftBuf, nRank);    // Perform direct FFT
-                        pFunc(pObject, pSubject, pFftBuf, nRank);           // Call the function
-                        dsp::packed_reverse_fft(pFftBuf, pFftBuf, nRank);   // Perform reverse FFT
-                        dsp::pcomplex_c2r(pFftBuf, pFftBuf, buf_size);      // Unpack complex numbers
+                        dsp::mul3(&pFftBuf[buf_size], pInBuf, pWnd, buf_size);      // Apply cosine window before transform
+                        dsp::pcomplex_r2c(pFftBuf, &pFftBuf[buf_size], buf_size);   // Convert from real to packed complex
+                        dsp::packed_direct_fft(pFftBuf, pFftBuf, nRank);            // Perform direct FFT
+                        pFunc(pObject, pSubject, pFftBuf, nRank);                   // Call the function
+                        dsp::packed_reverse_fft(pFftBuf, pFftBuf, nRank);           // Perform reverse FFT
+                        dsp::pcomplex_c2r(pFftBuf, pFftBuf, buf_size);              // Unpack complex numbers
                     }
                     else
-                        dsp::move(pFftBuf, pInBuf, buf_size);               // Copy data to FFT buffer
+                        dsp::mul3(pFftBuf, pInBuf, pWnd, buf_size);                 // Copy data to FFT buffer
 
                     // Apply signal to buffer
                     dsp::move(pOutBuf, &pOutBuf[frame_size], frame_size);   // Shift output buffer
                     dsp::fill_zero(&pOutBuf[frame_size], frame_size);       // Fill tail of input buffer with zeros
-                    dsp::fmadd3(pOutBuf, pFftBuf, pWnd, buf_size);          // Apply window and add to the output buffer
+                    dsp::fmadd3(pOutBuf, pFftBuf, pWnd, buf_size);          // Apply cosine window (-> squared cosine) and add to the output buffer
 
                     // Shift input buffer
                     dsp::move(pInBuf, &pInBuf[frame_size], frame_size);     // Shift input buffer
@@ -214,9 +215,10 @@ namespace lsp
                     if (pFunc != NULL)
                     {
                         // Perform FFT and processing
-                        dsp::pcomplex_r2c(pFftBuf, pInBuf, buf_size);       // Convert from real to packed complex
-                        dsp::packed_direct_fft(pFftBuf, pFftBuf, nRank);    // Perform direct FFT
-                        pFunc(pObject, pSubject, pFftBuf, nRank);           // Call the function
+                        dsp::mul3(&pFftBuf[buf_size], pInBuf, pWnd, buf_size);      // Apply cosine window before transform
+                        dsp::pcomplex_r2c(pFftBuf, &pFftBuf[buf_size], buf_size);   // Convert from real to packed complex
+                        dsp::packed_direct_fft(pFftBuf, pFftBuf, nRank);            // Perform direct FFT
+                        pFunc(pObject, pSubject, pFftBuf, nRank);                   // Call the function
                     }
 
                     // Apply signal to buffer