14.4.2 Off-line sound source separation (with postprocessing by HRLE)

A sample network file that separates sound by GHDSS node and post process for speech enhancement using HRLE is introduced here.

\includegraphics[width=0.8\textwidth ]{fig/recipes/LearningHARK-separation-hrle.png}
Figure 14.13: MAIN_LOOP (iterator)

Run the demo.sh with commandline arguments offline and HRLE. THen, you will find the separated files in sep_files/ directory. The file name is offlineHRLE_%d.wav

Figure 14.13 shows sample network of demoOfflineHRLE.n, which is off-line sound source separation with postprocessing by HRLE . The audio waveforms read from the files in the AudioStreamFromWave node are analyzed in MultiFFT , separated in GHDSS , postprocessed, synthesized in Synthesize and the audio waveforms are saved in SaveWavePCM . The postprocessing is realized by combinations of HRLE , EstimateLeak , CalcSpecAddPower , CalcSpecSubGain and SpectralGainFilter . Interference from a non-purpose sound is estimated from nondirectional noise and the sound sources detected and spectral levels for each band are adjusted.