Easily extensible unix software for spectral analysis, display modification, and synthesis of...

download Easily extensible unix software for spectral analysis, display modification, and synthesis of musical sounds James W. Beauchamp School of Music Dept.

If you can't read please download the document

Transcript of Easily extensible unix software for spectral analysis, display modification, and synthesis of...

  • Slide 1

Slide 2 Easily extensible unix software for spectral analysis, display modification, and synthesis of musical sounds James W. Beauchamp School of Music Dept. of Electrical and Computer Engineering University of Illinois at Urbana-Champaign [email protected] http://www.staff.uiuc.edu/~j-beauch Slide 3 Talk Topics SNDAN features time-domain utilities phase vocoder harmonic spectrum analysis spectral graphics, modification, resynthesis frequency tracking analysis, graphics, synthesis pitch detection, conversion to harmonic format SNDAN applications Future developments Conclusions Slide 4 SNDAN OVERVIEW: ANALYSIS FRONT END signal GRAPHICSMODIFICATIONRESYNTHESIS analysis data Slide 5 SNDAN OVERVIEW: SIGNAL VIEWING AND EDITING Slide 6 SNDAN OVERVIEW: SPECTRUM ANALYSIS Slide 7 SNDAN OVERVIEW: SPECTRUM VIEWING, MODIFICATION, AND RESYNTHESIS Slide 8 Phase Vocoder Analysis s(t) HAMMING WINDOW (double period) DISCRETE FOURIER TRANSFORM (FFT) w(t)s(t) OVERLAP BY 1/2 PERIOD THROW AWAY K/2, K ODD COMPUTE AMPLITUDES AND PHASES COMPUTE FREQUENCY DEVIATIONS harmonic data BANDLIMITED INTERPOLATION RESAMPLER s(t)s(t) f s (sample frequency) f a (analysis frequency) sound signal Slide 9 Harmonic Data Graphics EPS graphics display harmonic data 1D: Amplitude vs. Frequency (snapshot bar, line; comp. overlay) 1D & 2D: Frequency vs. Time (individual, spectrogram) 3D: Amplitude vs. Frequency and Time Inharmonicity vs. Time Spectral Centroid vs. Time Spectral Centroid vs. RMS Ampl. Spectral Irreg. vs. Time Inverse Spectral Density vs. Time Musical Pitch vs. Time Slide 10 Example 2D graph Slide 11 Example 3D graph Slide 12 Harmonic Data Modification harmonic data Smooth A k vs. time (t) Make A k (t) proportional to A rms (t) Smooth A k vs. frequency (k) Scale A k by k p to achieve new average centroid Scale A k to achieve designated spectrum or aux. spectrum aux. harmonic data Warp attack time Reduce duration without affecting attack and decay. Smooth f k vs. time (t) Make all f k (t) harmonic to f ave (t) Flatten f k to average or harmonic Quantize fund. freq. to ET pitch Slide 13 Harmonic Data Resynthesis harmonic data synthetic signal AMPLITUDE & FREQUENCY LINEAR INTERPOLATION (scale amplitude, freq, duration) AMPLITUDE & PHASE QUADRATIC INTERPOLATION Slide 14 Signal Modification Example Original Flute Time-smoothed Amplitudes Time-smoothed Amplitudes & Frequencies Time-smoothed Amplitudes & Flattened Frequencies Time-smoothed, RMSed, & Spectrum Envelope Smoothed & Flattened Frequencies Time-smoothed, Spectrum Envelope Smoothed & Flattened Frequencies Slide 15 Frequency Tracking (MQ) Data Analysis KAISER WINDOW WITH 100% ZERO FILL s(t)s(t) f s f min sound signal FFT WITH TYPICAL 6 MS HOP APPLY THRESHOLD IDENTIFY AND SAVE SPECTRAL PEAKS AT EACH FRAME COMPUTE EACH PEAKS AMPLITUDE, FREQUENCY AND PHASE partial (MQ) data A thresh CONNECT PEAKS TO NEXT FRAME PEAKS (TRACKS) Slide 16 Graphics partial data 2D: FREQUENCY VS. TIME 3D: AMPLITUDE VS. FREQUENCY VS. TIME EPS display Synthesis synthetic signal INTERPOLATION: AMPLITUDE - LINEAR PHASE - CUBIC partial data time scale Partial Data Processing Slide 17 Pitch Detection TWO-WAY MISMATCH HARMONIC MATCHING METHOD fundamental frequency data F 0 (t) f min f max partial data HARMONIC SIEVE fund freq data F 0 (t) harmonic data n har (no. of harmonics) frequency tolerance Harmonic Separation Slide 18 Saxophone Solo Pitch Plot:2D Peak Track Plot: Original SoundSynthesized from Partial Data Synthesized from Harmonic-reduced Data Partial Data Stretched x2 Harmonic Data Stretched x2 Harmonic Data Stretched x2 Smoothed Freq Slide 19 Applications Synthesis instrument development nonlinear and frequency modulation wavetable trumpet and piano Timbre investigations simplified sounds for discrimination studies normalized sounds for MDS studies perturbed sounds for discrimination studies synthesis quality evaluation Music composition using Music 4C Slide 20 Future Developments More features for partial data format Integrate programs into single program More advanced analysis front end Multi-Channel Create GUI interface Real time Port to more platforms Slide 21 SNDAN Conclusions Provides analysis, graphics, modification, and synthesis Specialized for musical sounds Two spectrum data formats: harmonic and partial Contains pitch detector Unix source code modular and easily extensible Source code available at: http://www.staff.uiuc.edu/~j-beauch/software/sndan/ DOS binary version available at: http://ftp.cs.bath.ac.uk/pub/dream/SNDAN32/ Real-time GUI spinoff analyzer for Mac available at: http://www.staff.uiuc.edu/~j-beauch/software/armadillo/