180 likes | 453 Views
Analysis/Synthesis of Transients in Audio Signals. François Xavier Nsabimana. Outline. Signal representation Sinusoidal Modeling: SM Spectral Modeling Synthesis: SMS SM: Sinusoidal Modeling RA: Residual Approximation Transient Modeling Synthesis: TMS Sinusoids Modeling
E N D
Analysis/Synthesis of Transients in Audio Signals François Xavier Nsabimana
Outline • Signal representation • Sinusoidal Modeling: SM • Spectral Modeling Synthesis: SMS • SM: Sinusoidal Modeling • RA: Residual Approximation • Transient Modeling Synthesis: TMS • Sinusoids Modeling • Transient Modeling • Noise Modeling Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Signal representation Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Sinusoidal Modeling • Signal as sum of sinusoids: • Spectrogram: sinusoids presented as horizontal lines Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Spectral Modeling Synthesis (1) • SMS: Deterministic plus stochastic model (Serra) • RA: Residual approximation • spectral envelope • Filter from envelope • White noise filtering Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Transient Modeling Synthesis (1) • Signal composed of : • Sinusoids • Transients • Noise. • Characteristic of transient: • Impulsive (in time domain) • spectrogram: Vertical lines Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Transient Modeling Synthesis (2) • TMS approach (Verma): • time/frequency duality • Sinusoids Impulsive • Impulsive Sinusoids • Transformation to use: • Discrete Cosine Transform (DCT) • Mapping transients into sinusoids • High frequency sinusoids • Low frequency sinusoids Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Transient Modeling Synthesis (3) • TMS algorithm: • Nonoverlapping blocks • DCT block length (1sec) • Transients look like small entities within the block • SM of DCT signal: • 30 - 60 SM frames per DCT block • transients well modeled • About 10 SM frames per DCT block • transients only located Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Transient Modeling Synthesis (4) • Test of TMS model with synthetic signal • Drawback: • transients at the end of DCT block not detected Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Test of TMS model with real signal Drawback: transients at the end of DCT block still not detected Transient Modeling Synthesis (5) Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Transient Modeling Synthesis (6) • Improvement of TMS : • Increase of Resolution • Zero-padding of every signal block before performing DCT • Zero-padding length: • Sufficient up half the DCT block length • Transients at the end of DCT block : • All detected Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Transient Modeling Synthesis (7) • Application of TMS to real signal: • Castanet sound file • SM block length: 1024 • About 43 SM frames per DCT block. • Traces of transients in the residual. Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Transient Modeling Synthesis (8) • SM block length: 512 • About 86 SM frames per DCT block . • Traces of transients less visible in the residual • Small windows: • Good for time resolution Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Sound Results with TMS (1) • Castanet sound file: • Original : • STN : • TMS : • TM : Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Sound Results with TMS (2) • Drums sound file: • Original : • STN : • TMS : • TM : Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Conclusion • Improvement of Spectral Modeling Synthesis • Implementation and Improvement of TMS model • Application of TMS model to percussion music signals Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals
Future work. • Implementation of a more robust method for the separation of transients and noise before modeling these two components. • Use of Linear Prediction to locate transient area • Use of Envelope Detection to build the threshold function Dipl.-Ing. Francois Nsabimana - Analysis/Synthesis of Transients in Audio Signals