diff options
| -rw-r--r-- | __init__.py | 4 | ||||
| -rw-r--r-- | basetypes.py | 21 | ||||
| -rw-r--r-- | mq.py | 60 | ||||
| -rw-r--r-- | readme.txt | 3 | ||||
| -rw-r--r-- | sms.py | 5 | ||||
| -rw-r--r-- | sms/sineSynth.c | 2 | ||||
| -rw-r--r-- | sndobj/SinSyn.cpp | 394 |
7 files changed, 283 insertions, 206 deletions
diff --git a/__init__.py b/__init__.py index 44f2c37..cd46097 100644 --- a/__init__.py +++ b/__init__.py @@ -18,10 +18,10 @@ from basetypes import Peak, PeakDetection, Partial, PartialTracking, Synthesis, from basetypes import compare_peak_amps, compare_peak_freqs from sndobj import SndObjPeakDetection, SndObjPartialTracking, SndObjSynthesis from sms import SMSPeakDetection, SMSPartialTracking, SMSSynthesis, SMSResidual -from mq import MQPeakDetection, MQPartialTracking +from mq import MQPeakDetection, MQPartialTracking, MQSynthesis +from plot import plot_peaks, plot_partials import numpy -#float = numpy.float64 def array (n, type=float): return(numpy.array(n, dtype=type)) diff --git a/basetypes.py b/basetypes.py index b2f0ae8..28311ad 100644 --- a/basetypes.py +++ b/basetypes.py @@ -267,15 +267,29 @@ class PartialTracking(object): class Synthesis(object): "Synthesise audio from spectral analysis data" def __init__(self): + self._frame_size = 512 self._hop_size = 512 self._max_partials = 100 + self._sampling_rate = 44100 # properties + frame_size = property(lambda self: self.get_frame_size(), + lambda self, x: self.set_frame_size(x)) hop_size = property(lambda self: self.get_hop_size(), lambda self, x: self.set_hop_size(x)) max_partials = property(lambda self: self.get_max_partials(), lambda self, x: self.set_max_partials(x)) + max_partials = property(lambda self: self.get_max_partials(), + lambda self, x: self.set_max_partials(x)) + sampling_rate = property(lambda self: self.get_sampling_rate(), + lambda self, x: self.set_sampling_rate(x)) + def get_frame_size(self): + return self._frame_size + + def set_frame_size(self, frame_size): + self._frame_size = frame_size + def get_hop_size(self): return self._hop_size @@ -288,6 +302,12 @@ class Synthesis(object): def set_max_partials(self, num_partials): self._max_partials = num_partials + def get_sampling_rate(self): + return self._sampling_rate + + def set_sampling_rate(self, sampling_rate): + self._sampling_rate = sampling_rate + def synth_frame(self, partials): "Synthesises a frame of audio, given a list of peaks from tracks" raise Exception("NotYetImplemented") @@ -336,3 +356,4 @@ class Residual(object): def synth(self, synth, original): "Calculate and return a synthesised residual signal" raise Exception("NotYetImplemented") + @@ -63,7 +63,7 @@ class MQPeakDetection(simpl.PeakDetection): p = simpl.Peak() p.amplitude = current_mag p.frequency = (bin - 1) * self._fundamental - p.phase = np.angle(spectrum[bin-1]) + p.phase = np.angle(f[bin-1]) current_peaks.append(p) prev_mag = current_mag current_mag = next_mag @@ -147,7 +147,7 @@ class MQPartialTracking(simpl.PartialTracking): for peak in self._current_frame: if peak.is_start_of_partial(): partial = simpl.Partial() - partial.starting_frame = frame_number-1 + partial.starting_frame = frame_number partial.add_peak(peak) self.partials.append(partial) match = self._find_closest_match(peak, frame) @@ -179,7 +179,7 @@ class MQPartialTracking(simpl.PartialTracking): # create a new track by adding a peak in the current frame, matched to p, # with amplitude 0 partial = simpl.Partial() - partial.starting_frame = frame_number-1 + partial.starting_frame = frame_number new_peak = simpl.Peak() new_peak.amplitude = 0 new_peak.frequency = p.frequency @@ -189,3 +189,57 @@ class MQPartialTracking(simpl.PartialTracking): self._current_frame = frame return frame_partials + +class MQSynthesis(simpl.Synthesis): + def __init__(self): + simpl.Synthesis.__init__(self) + self._current_frame = simpl.zeros(self.frame_size) + + def hz_to_radians(self, frequency): + if not frequency: + return 0.0 + else: + return (frequency * 2.0 * np.pi) / self.sampling_rate + + def synth_frame(self, peaks): + "Synthesises a frame of audio, given a list of peaks from tracks" + self._current_frame *= 0.0 + + for p in peaks: + # get values for last amplitude, frequency and phase + # these are the initial values of the instantaneous amplitude/frequency/phase + current_freq = self.hz_to_radians(p.frequency) + if not p.previous_peak: + prev_amp = 0.0 + prev_freq = current_freq + prev_phase = p.phase - (current_freq * self.frame_size) + while prev_phase >= np.pi: prev_phase -= 2.0 * np.pi + while prev_phase < -np.pi: prev_phase += 2.0 * np.pi + else: + prev_amp = p.previous_peak.amplitude + prev_freq = self.hz_to_radians(p.previous_peak.frequency) + prev_phase = p.previous_peak.phase + + # amplitudes are linearly interpolated between frames + inst_amp = prev_amp + amp_inc = (p.amplitude - prev_amp) / self.frame_size + + # freqs/phases are calculated by cubic interpolation + freq_diff = current_freq - prev_freq + x = ((prev_phase + (prev_freq * self.frame_size) - p.phase) + + (freq_diff * (self.frame_size / 2.0))) + x /= (2.0 * np.pi) + m = int(np.round(x)) + phase_diff = p.phase - prev_phase - (prev_freq * self.frame_size) + (2.0 * np.pi * m) + alpha = ((3.0 / (self.frame_size**2)) * phase_diff) - (freq_diff / self.frame_size) + beta = ((-2.0 / (self.frame_size**3)) * phase_diff) + (freq_diff / (self.frame_size**2)) + + # calculate output samples + for i in range(self.frame_size): + inst_amp += amp_inc + inst_phase = prev_phase + (prev_freq * i) + (alpha * (i**2)) + (beta * (i**3)) + #print inst_phase + self._current_frame[i] += (2.0 * inst_amp) * np.cos(inst_phase) + + return self._current_frame + @@ -116,4 +116,5 @@ sms: mq: - fix peak detection module - window size should change depending on detected pitch. No pitch detection algorithm described in the - original paper, so can use the SMS algorithm for now.
\ No newline at end of file + original paper, so can use the SMS algorithm for now. + @@ -282,6 +282,8 @@ class SMSSynthesis(simpl.Synthesis): pysms.sms_init() self._synth_params = pysms.SMS_SynthParams() self._synth_params.iDetSynthType = pysms.SMS_DET_SIN + # use the default simpl hop size instead of the default SMS hop size + self._synth_params.sizeHop = self._hop_size pysms.sms_initSynth(self._synth_params) self._current_frame = simpl.zeros(self.hop_size) self._analysis_frame = pysms.SMS_Data() @@ -295,8 +297,6 @@ class SMSSynthesis(simpl.Synthesis): SMSSynthesis._instances -= 1 # properties - sampling_rate = property(lambda self: self.get_sampling_rate(), - lambda self, x: self.set_sampling_rate(x)) synthesis_type = property(lambda self: self.get_synthesis_type(), lambda self, x: self.set_synthesis_type(x)) num_stochastic_coeffs = property(lambda self: self.get_num_stochastic_coeffs(), @@ -402,3 +402,4 @@ class SMSResidual(simpl.Residual): if pysms.sms_findResidual(synth, original, residual, self._analysis_params) == -1: raise Exception("Residual error: Synthesised audio and original audio have different lengths") return residual + diff --git a/sms/sineSynth.c b/sms/sineSynth.c index aaca41d..58768bd 100644 --- a/sms/sineSynth.c +++ b/sms/sineSynth.c @@ -59,7 +59,7 @@ static void SinePhaSynth (sfloat fFreq, sfloat fMag, sfloat fPhase, (pLastFrame->pFSinFreq[iTrack] * sizeBuffer); fPhase = fTmp - floor(fTmp / TWO_PI) * TWO_PI; } - + /* caculate the instantaneous amplitude */ fMagIncr = (fMag - pLastFrame->pFSinAmp[iTrack]) / sizeBuffer; fInstMag = pLastFrame->pFSinAmp[iTrack]; diff --git a/sndobj/SinSyn.cpp b/sndobj/SinSyn.cpp index 7452b95..7d31e15 100644 --- a/sndobj/SinSyn.cpp +++ b/sndobj/SinSyn.cpp @@ -1,4 +1,4 @@ - + //////////////////////////////////////////////////////////////////////// // This file is part of the SndObj library // @@ -24,86 +24,86 @@ SinSyn::SinSyn(){ - m_factor = m_vecsize/m_sr; - m_facsqr = m_factor*m_factor; - m_ptable = 0; - m_size = 0; - m_LoTWOPI = 0.f; - m_maxtracks = 0; - m_freqs = 0; - m_amps = 0; - m_phases = 0; - m_trackID = 0; - m_scale = 0.f; - m_ratio = 0.f; - m_tracks = 0; - AddMsg("max tracks", 21); - AddMsg("scale", 23); - AddMsg("table", 24); + m_factor = m_vecsize/m_sr; + m_facsqr = m_factor*m_factor; + m_ptable = 0; + m_size = 0; + m_LoTWOPI = 0.f; + m_maxtracks = 0; + m_freqs = 0; + m_amps = 0; + m_phases = 0; + m_trackID = 0; + m_scale = 0.f; + m_ratio = 0.f; + m_tracks = 0; + AddMsg("max tracks", 21); + AddMsg("scale", 23); + AddMsg("table", 24); } SinSyn::SinSyn(SinAnal* input, int maxtracks, Table* table, - double scale, int vecsize, double sr) - :SndObj(input, vecsize, sr){ - - m_ptable = table; - m_size = m_ptable->GetLen(); - m_LoTWOPI = m_size/TWOPI; - - m_factor = m_vecsize/m_sr; - m_facsqr = m_factor*m_factor; - m_maxtracks = maxtracks; - m_tracks = 0; - m_scale = scale; - m_input = input; - m_freqs = new double[m_maxtracks]; - m_amps = new double[m_maxtracks]; - m_phases = new double[m_maxtracks]; - m_trackID = new int[m_maxtracks]; - - memset(m_phases, 0, sizeof(double)*m_maxtracks); - - m_incr = 0.f; - m_ratio = m_size/m_sr; - AddMsg("max tracks", 21); - AddMsg("scale", 23); - AddMsg("table", 24); - AddMsg("timescale", 24); - memset(m_trackID, 0, sizeof(int)); + double scale, int vecsize, double sr) +:SndObj(input, vecsize, sr){ + + m_ptable = table; + m_size = m_ptable->GetLen(); + m_LoTWOPI = m_size/TWOPI; + + m_factor = m_vecsize/m_sr; + m_facsqr = m_factor*m_factor; + m_maxtracks = maxtracks; + m_tracks = 0; + m_scale = scale; + m_input = input; + m_freqs = new double[m_maxtracks]; + m_amps = new double[m_maxtracks]; + m_phases = new double[m_maxtracks]; + m_trackID = new int[m_maxtracks]; + + memset(m_phases, 0, sizeof(double)*m_maxtracks); + + m_incr = 0.f; + m_ratio = m_size/m_sr; + AddMsg("max tracks", 21); + AddMsg("scale", 23); + AddMsg("table", 24); + AddMsg("timescale", 24); + memset(m_trackID, 0, sizeof(int)); } SinSyn::~SinSyn(){ - delete[] m_freqs; - delete[] m_amps; - delete[] m_phases; - delete[] m_trackID; + delete[] m_freqs; + delete[] m_amps; + delete[] m_phases; + delete[] m_trackID; } -void + void SinSyn::SetTable(Table *table) { - m_ptable = table; - m_size = m_ptable->GetLen(); - m_LoTWOPI = m_size/TWOPI; - m_ratio = m_size/m_sr; + m_ptable = table; + m_size = m_ptable->GetLen(); + m_LoTWOPI = m_size/TWOPI; + m_ratio = m_size/m_sr; } int SinSyn::Connect(char* mess, void* input){ - switch (FindMsg(mess)){ + switch (FindMsg(mess)){ + + case 24: + SetTable((Table *) input); + return 1; - case 24: - SetTable((Table *) input); - return 1; + default: + return SndObj::Connect(mess,input); - default: - return SndObj::Connect(mess,input); - - } + } } @@ -111,158 +111,158 @@ SinSyn::Connect(char* mess, void* input){ int SinSyn::Set(char* mess, double value){ - switch(FindMsg(mess)){ + switch(FindMsg(mess)){ - case 21: - SetMaxTracks((int)value); - return 1; + case 21: + SetMaxTracks((int)value); + return 1; - case 23: - SetScale(value); - return 1; - - default: - return SndObj::Set(mess, value); + case 23: + SetScale(value); + return 1; - } + default: + return SndObj::Set(mess, value); + + } } void SinSyn::SetMaxTracks(int maxtracks){ - if(m_maxtracks){ + if(m_maxtracks){ - delete[] m_freqs; - delete[] m_amps; - delete[] m_phases; - delete[] m_trackID; + delete[] m_freqs; + delete[] m_amps; + delete[] m_phases; + delete[] m_trackID; - } + } - m_maxtracks = maxtracks; - m_freqs = new double[m_maxtracks]; - m_amps = new double[m_maxtracks]; - m_phases = new double[m_maxtracks]; - m_trackID = new int[m_maxtracks]; + m_maxtracks = maxtracks; + m_freqs = new double[m_maxtracks]; + m_amps = new double[m_maxtracks]; + m_phases = new double[m_maxtracks]; + m_trackID = new int[m_maxtracks]; } short SinSyn::DoProcess() { - - if(m_input){ - - double ampnext,amp,freq, freqnext, phase,phasenext; - double a2, a3, phasediff, cph; - int i3, i, j, ID, track; - int notcontin = 0; - bool contin = false; - int oldtracks = m_tracks; - double* tab = m_ptable->GetTable(); - if((m_tracks = ((SinAnal *)m_input)->GetTracks()) > - m_maxtracks) m_tracks = m_maxtracks; - - memset(m_output, 0, sizeof(double)*m_vecsize); - - // for each track - i = j = 0; - while(i < m_tracks*3){ - i3 = i/3; - ampnext = m_input->Output(i)*m_scale; - freqnext = m_input->Output(i+1)*TWOPI; - phasenext = m_input->Output(i+2); - ID = ((SinAnal *)m_input)->GetTrackID(i3); - - j = i3+notcontin; - - if(i3 < oldtracks-notcontin){ - - if(m_trackID[j]==ID){ - // if this is a continuing track - track = j; - contin = true; - freq = m_freqs[track]; - phase = m_phases[track]; - amp = m_amps[track]; - - } - else { - // if this is a dead track - contin = false; - track = j; - freqnext = freq = m_freqs[track]; - phase = m_phases[track]; - phasenext = phase + freq*m_factor; - amp = m_amps[track]; - ampnext = 0.f; - } - } - - else{ - // new tracks - contin = true; - track = -1; - freq = freqnext; - phase = phasenext - freq*m_factor; - amp = 0.f; - } - - // phasediff - phasediff = phasenext - phase; - while(phasediff >= PI) phasediff -= TWOPI; - while(phasediff < -PI) phasediff += TWOPI; - // update phasediff to match the freq - cph = ((freq+freqnext)*m_factor/2. - phasediff)/TWOPI; - phasediff += TWOPI * Ftoi(cph + 0.5); - // interpolation coefs - a2 = 3./m_facsqr * (phasediff - m_factor/3.*(2*freq+freqnext)); - a3 = 1./(3*m_facsqr) * (freqnext - freq - 2*a2*m_factor); - - // interpolation resynthesis loop - double inc1, inc2, a, ph, cnt, frac; - int ndx; - a = amp; - ph = phase; - cnt = 0; - inc1 = (ampnext - amp)/m_vecsize; - inc2 = 1/m_sr; - for(m_vecpos=0; m_vecpos < m_vecsize; m_vecpos++){ - - if(m_enable) { - // table lookup oscillator - ph *= m_LoTWOPI; - while(ph < 0) ph += m_size; - while(ph >= m_size) ph -= m_size; - ndx = Ftoi(ph); - frac = ph - ndx; - m_output[m_vecpos] += a*(tab[ndx] + (tab[ndx+1] - tab[ndx])*frac); - a += inc1; - cnt += inc2; - ph = phase + cnt*(freq + cnt*(a2 + a3*cnt)); - - } - else m_output[m_vecpos] = 0.f; - } - - // keep amp, freq, and phase values for next time - if(contin){ - m_amps[i3] = ampnext; - m_freqs[i3] = freqnext; - while(phasenext < 0) phasenext += TWOPI; - while(phasenext >= TWOPI) phasenext -= TWOPI; - m_phases[i3] = phasenext; - m_trackID[i3] = ID; - i += 3; - } else notcontin++; - } - - return 1; - } - else { - m_error = 1; - return 0; - } - + + if(m_input){ + + double ampnext,amp,freq, freqnext, phase,phasenext; + double a2, a3, phasediff, cph; + int i3, i, j, ID, track; + int notcontin = 0; + bool contin = false; + int oldtracks = m_tracks; + double* tab = m_ptable->GetTable(); + if((m_tracks = ((SinAnal *)m_input)->GetTracks()) > + m_maxtracks) m_tracks = m_maxtracks; + + memset(m_output, 0, sizeof(double)*m_vecsize); + + // for each track + i = j = 0; + while(i < m_tracks*3){ + i3 = i/3; + ampnext = m_input->Output(i)*m_scale; + freqnext = m_input->Output(i+1)*TWOPI; + phasenext = m_input->Output(i+2); + ID = ((SinAnal *)m_input)->GetTrackID(i3); + + j = i3+notcontin; + + if(i3 < oldtracks-notcontin){ + + if(m_trackID[j]==ID){ + // if this is a continuing track + track = j; + contin = true; + freq = m_freqs[track]; + phase = m_phases[track]; + amp = m_amps[track]; + + } + else { + // if this is a dead track + contin = false; + track = j; + freqnext = freq = m_freqs[track]; + phase = m_phases[track]; + phasenext = phase + freq*m_factor; + amp = m_amps[track]; + ampnext = 0.f; + } + } + + else{ + // new tracks + contin = true; + track = -1; + freq = freqnext; + phase = phasenext - freq*m_factor; + amp = 0.f; + } + + // phasediff + phasediff = phasenext - phase; + while(phasediff >= PI) phasediff -= TWOPI; + while(phasediff < -PI) phasediff += TWOPI; + // update phasediff to match the freq + cph = ((freq+freqnext)*m_factor/2. - phasediff)/TWOPI; + phasediff += TWOPI * Ftoi(cph + 0.5); + // interpolation coefs + a2 = 3./m_facsqr * (phasediff - m_factor/3.*(2*freq+freqnext)); + a3 = 1./(3*m_facsqr) * (freqnext - freq - 2*a2*m_factor); + + // interpolation resynthesis loop + double inc1, inc2, a, ph, cnt, frac; + int ndx; + a = amp; + ph = phase; + cnt = 0; + inc1 = (ampnext - amp)/m_vecsize; + inc2 = 1/m_sr; + for(m_vecpos=0; m_vecpos < m_vecsize; m_vecpos++){ + + if(m_enable) { + // table lookup oscillator + ph *= m_LoTWOPI; + while(ph < 0) ph += m_size; + while(ph >= m_size) ph -= m_size; + ndx = Ftoi(ph); + frac = ph - ndx; + m_output[m_vecpos] += a*(tab[ndx] + (tab[ndx+1] - tab[ndx])*frac); + a += inc1; + cnt += inc2; + ph = phase + cnt*(freq + cnt*(a2 + a3*cnt)); + + } + else m_output[m_vecpos] = 0.f; + } + + // keep amp, freq, and phase values for next time + if(contin){ + m_amps[i3] = ampnext; + m_freqs[i3] = freqnext; + while(phasenext < 0) phasenext += TWOPI; + while(phasenext >= TWOPI) phasenext -= TWOPI; + m_phases[i3] = phasenext; + m_trackID[i3] = ID; + i += 3; + } else notcontin++; + } + + return 1; + } + else { + m_error = 1; + return 0; + } + } |