DAVID A JAFFE Ten Criteria for Evaluating synthesis Techniques part1

10 EVALUATION CRITERIA: PARAMETRIC 1 - 5
1. how INTUITIVE are the parameters
  1. STRONG -MAPPING control parameters to musical parameters
    1. EGS STRONG
      1. dynamics
      2. articulation
      3. ANALOG SYNTHESIS LPF maps BW / AMP less brightness with softer dynamics
        LOW VALUES = PP
        softer notes = lower AMP + less high freq
        HI VALUES = FF
        louder notes = higher AMP + more high freq
        physical nonlinearities become more prominant as instr are played louder
        ADJUSTABLE ON note-by-note basis: DECOUPLES
        DISTANCE - FUNCTION OF LOUDNESS
        DYNAMICS - FUNCTION OF BRIGHTNESS
  2. WEAK UNMAPPED mathematical variables
    1. EGS WEAK
      1. Changing FM INDEX OF A CASCADE MODULATOR SLIGHTLY
        DRASTIC, DIFFICULT TO PREDICT CHANGE IN TONAL QUALITY EG FROM DRUYM TO WOODBLOCK
        Annoying for composers
        Problematic for performer - in live situ
2. how PERCEPTIBLE are parameter changes
  1. more parameters - weaker the impact of each
  2. STRONG
    1. sig. ch. = obvious audible effect
    2. too strong = uncontrollable
    3. e.g. a low pass filter proportional to cutoff freq (expon or linear)
  3. WEAK
    1. sig. ch. = barely audible
    2. too weak = fruitless arbitrary alterations
    3. METAPARAMETERS
      1. EG1a
        Used to enable more intuitive and perceptible changes to synthesis methods - e.g. brightness to each array element freq, AMP array, brightness can be applied across each element of the array
        ActualAmps[i]=Amps[i]*MAX(brightness, 0.001)^log2(MAX(Freqs[i],100.0)/50.0)
        addSyn used as inv FFT the LPF - log2 and MAX() rtns max of its args,
      2. EG1b
        addSyn * LineSeg - multidimentional timbral space
        values added to scale addSyn partials producing complex interpolations (mcNabb 8f1)
      3. EG2
        Digital Sampling (*ID synth tech). edit 1 sample no effect used wtih HYBRID PARAMETERIZED PROCESSING MODULES eg filters, amp envs etc
    4. EG1 addSyn - ch. amp of each partial could be weak for hi harmonics
3. how PHYSICAL are parameters
  1. cntrls synthetic instr. in same way as real-world instr. is cntrled
    1. EG1
      1. Violin - waveguide model bow pressure/velocity (smith 92)
    2. EG2
      1. Clarinet mouth breath pressure parameter (cook 88)
  2. ADV
    1. 1 good at: Complx behaviours at unstable moments in a tones evolution eg attack and transistions
    2. 2 no need for const. analytical modelling/database player/composer/implemntor neednt understand what happens in output wav during attack/trams te physical nature of parameter cuases the correct result
  3. DIS
    1. For nonexpert players - nonphysical INTUITIVE cntrls - more appropr.
  4. for non-phys. synth parameters (eg. amp of addSyn partial -
    1. SOLUTION - MAP WITH META- PARAMETER
    2. PROB - need thorough ACOUSTICAL KNOWLEDGE to defne mapping adquately
4. how WELL BEHAVED are the parameters
  1. EGS - LINEAR TECHNIQUES
    1. when CNTRL RATE = SLOW RELATIVE TO AUDIO SR
      1. STRONG
        STRONG - PROPORTIONAL change to sound wth cntrler
        FILTERS - GOOD DURING TRANSITIONS
        PERRY COOK WAVEGUIDE-BASED VOCAL SYNTH MODEL 90
        MODELS vocal tract - series waveguide filters
        Each waveguide filter = xsect. of vocal tract
        each vowel = physical interpretation - 1 or sev section diameters: shrink / incr.
        interpolating between vowels corres. dir. to real-world spatial interpolation
        Each intermediate val. have INTUITIVE PHYSICAL interpretation = WELLBEHAVED
    2. CNTRL parameters change RAPIDLY - ENERGY IS INECTED INTO TH SSYSTEM WTH POSS OF UNPREDICTABLE UNEXPECTED RESULTS
      1. WEAK
        WEAK - small parameter ch - WILD, UNPREDICTABLE sonic outcome
        EG 3 NONLINEAR FB TECH (Mcintyre/ woodhouse 83
        nonlin. FB in phys. mod. wind instr.
        EXTREME SENS. to INITIAL COND.
        DEF. OF CHAOTIC SYS.
        can be modelled in REGIONS of its STATE SPACE for PREDICABLE EFFECTIVE HAEVIOUR
      2. E.G. AUDIO RATE AMPLITUDE SCALING
        E.G. SINUSOIDAL
        AM - 2 sidebands are produced for each componente in original signal = NONLINEAR - SIDEBANDS ARE AT FREQ CORRESONDING TO THE FREQ OF EACH COMPONENT PLUS OR MINUS THE MOD FREQ
        E.G. NON- SINUSOIDAL
        EVEN MORE CMPLX than above
      3. EG2 FILTER STRUCTURES THAT BECOME UNSTABLE DURING TRANSITIONS FROM 1 SET OF COEFFICIENTS TO ANOTHER
5. how ROBUST is the sound's IDENTITY
  1. HOW WELL SOUND RETAINS ID IN CONTEXT OF VARIATION
    1. True rep. of expres.. instr. = FAMILY OF RELATED SOUNDS req.
    2. not repr. of 1 note
  2. Inst. Def. wth grt variety in context of particular sounrce ID
  3. synthesis NOT a snapshot = expressive OVER TIME
  4. EG Good
    1. PHYS MOD
      1. PhM Waveguide
        Karplus- Strong String
        rich expr. vocab. not loose ID
        e.g. pick pos, string: flex, thickness, dyamics, decay char
        ks 83 / jaffe J smith 83
    2. SMPLING Power / flexib. CAN BE grtly enhan.
      1. Multiple rec. xros
        BW
        dynamic contours
        attacks
        Vibrato
        max multiple guises
      2. filters = dynamics
      3. linear interpolation for timbral variety eg thru BW n.b. take care to MATCH PITCH AND PHASE
      4. as many samples per inst. as memory allows - more guises covered more expressive result
  5. EG POOR
    1. Pure Sampling
      1. add var. to one sample = 2D cardboard sound
        e.g. articulation, dynamics, timbre falls over if pused lightly in any direction
    2. 1 pitch
  6. sampling = challenge + opportunity
identity synthesis technique = DIGITAL SAMPLING. Each SAMPLE is a PARAMETER itself

Next up

DAVID A JAFFE Ten Criteria for Evaluating synthesis Techniques part1

Description

Resource summary

Similar

	Created by gpud over 9 years ago