tomwalters@12: // Copyright 2010, Thomas Walters
tomwalters@12: //
tomwalters@12: // AIM-C: A C++ implementation of the Auditory Image Model
tomwalters@12: // http://www.acousticscale.org/AIMC
tomwalters@12: //
tomwalters@45: // Licensed under the Apache License, Version 2.0 (the "License");
tomwalters@45: // you may not use this file except in compliance with the License.
tomwalters@45: // You may obtain a copy of the License at
tomwalters@12: //
tomwalters@45: //     http://www.apache.org/licenses/LICENSE-2.0
tomwalters@12: //
tomwalters@45: // Unless required by applicable law or agreed to in writing, software
tomwalters@45: // distributed under the License is distributed on an "AS IS" BASIS,
tomwalters@45: // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
tomwalters@45: // See the License for the specific language governing permissions and
tomwalters@45: // limitations under the License.
tomwalters@12: 
tomwalters@12: /*!
tomwalters@12:  * \author Thomas Walters <tom@acousticscale.org>
tomwalters@12:  * \date created 2010/02/19
tomwalters@12:  * \version \$Id$
tomwalters@12:  */
tomwalters@12: 
tomwalters@15: #include <cmath>
tomwalters@15: 
tomwalters@12: #include "Modules/SSI/ModuleSSI.h"
tomwalters@12: 
tomwalters@12: namespace aimc {
tomwalters@52: #ifdef _MSC_VER
tomwalters@52: // MSVC doesn't define log2()
tomwalters@52: float log2(float n) {
tomwalters@52:   return log(n) / log(2.0);  
tomwalters@52: }
tomwalters@52: #endif
tomwalters@52: 
tomwalters@12: ModuleSSI::ModuleSSI(Parameters *params) : Module(params) {
tomwalters@12:   module_description_ = "Size-shape image (aka the 'sscAI')";
tomwalters@12:   module_identifier_ = "ssi";
tomwalters@12:   module_type_ = "ssi";
tomwalters@12:   module_version_ = "$Id$";
tomwalters@12: 
tomwalters@32:   // Cut off the SSI at the end of the first cycle
tomwalters@32:   do_pitch_cutoff_ = parameters_->DefaultBool("ssi.pitch_cutoff", false);
tomwalters@32: 
tomwalters@32:   // Weight the values in each channel more strongly if the channel was
tomwalters@32:   // truncated due to the pitch cutoff. This ensures that the same amount of
tomwalters@32:   // energy remains in the SSI spectral profile
tomwalters@32:   weight_by_cutoff_ = parameters_->DefaultBool("ssi.weight_by_cutoff", false);
tomwalters@32: 
tomwalters@32:   // Weight the values in each channel more strongly if the channel was
tomwalters@32:   // scaled such that the end goes off the edge of the computed SSI.
tomwalters@32:   // Again, this ensures that the overall energy of the spectral profile
tomwalters@32:   // remains the same.
tomwalters@32:   weight_by_scaling_ = parameters_->DefaultBool("ssi.weight_by_scaling",
tomwalters@32:                                                 false);
tomwalters@32: 
tomwalters@32:   // Time from the zero-lag line of the SAI from which to start searching
tomwalters@32:   // for a maximum in the input SAI's temporal profile.
tomwalters@32:   pitch_search_start_ms_ = parameters_->DefaultFloat(
tomwalters@32:     "ssi.pitch_search_start_ms", 2.0f);
tomwalters@32: 
tomwalters@32:   // Total width in cycles of the whole SSI
tomwalters@32:   ssi_width_cycles_ = parameters_->DefaultFloat("ssi.width_cycles", 10.0f);
tomwalters@32: 
tomwalters@32:   // Set to true to make the cycles axis logarithmic (ie indexing by gamma
tomwalters@32:   // rather than by cycles) 
tomwalters@32:   log_cycles_axis_ = parameters_->DefaultBool("ssi.log_cycles_axis", true);
tomwalters@32: 
tomwalters@32:   // The centre frequency of the channel which will just fill the complete
tomwalters@32:   // width of the SSI buffer
tomwalters@32:   pivot_cf_ = parameters_->DefaultFloat("ssi.pivot_cf", 1000.0f);
tomwalters@12: }
tomwalters@12: 
tomwalters@12: ModuleSSI::~ModuleSSI() {
tomwalters@12: }
tomwalters@12: 
tomwalters@12: bool ModuleSSI::InitializeInternal(const SignalBank &input) {
tomwalters@12:   // Copy the parameters of the input signal bank into internal variables, so
tomwalters@12:   // that they can be checked later.
tomwalters@12:   sample_rate_ = input.sample_rate();
tomwalters@12:   buffer_length_ = input.buffer_length();
tomwalters@12:   channel_count_ = input.channel_count();
tomwalters@12: 
tomwalters@32:   ssi_width_samples_ = sample_rate_ * ssi_width_cycles_ / pivot_cf_;
tomwalters@15:   if (ssi_width_samples_ > buffer_length_) {
tomwalters@15:     ssi_width_samples_ = buffer_length_;
tomwalters@32:     float cycles = ssi_width_samples_ * pivot_cf_ / sample_rate_;
tomwalters@15:     LOG_INFO(_T("Requested SSI width of %f cycles is too long for the "
tomwalters@15:                 "input buffer length of %d samples. The SSI will be "
tomwalters@15:                 "truncated at %d samples wide. This corresponds to a width "
tomwalters@15:                 "of %f cycles."), ssi_width_cycles_, buffer_length_,
tomwalters@15:                 ssi_width_samples_, cycles);
tomwalters@15:     ssi_width_cycles_ = cycles;
tomwalters@15:   }
tomwalters@15:   output_.Initialize(channel_count_, ssi_width_samples_, sample_rate_);
tomwalters@12:   return true;
tomwalters@12: }
tomwalters@12: 
tomwalters@12: void ModuleSSI::ResetInternal() {
tomwalters@12: }
tomwalters@12: 
tomwalters@32: int ModuleSSI::ExtractPitchIndex(const SignalBank &input) const {
tomwalters@32:   // Generate temporal profile of the SAI
tomwalters@32:   vector<float> sai_temporal_profile(buffer_length_, 0.0f);
tomwalters@32:   for (int i = 0; i < buffer_length_; ++i) {
tomwalters@32:     float val = 0.0f;
tomwalters@32:     for (int ch = 0; ch < channel_count_; ++ch) {
tomwalters@32:       val += input.sample(ch, i);
tomwalters@32:     }
tomwalters@32:     sai_temporal_profile[i] = val;
tomwalters@32:   }
tomwalters@32: 
tomwalters@32:   // Find pitch value
tomwalters@32:   int start_sample = floor(pitch_search_start_ms_ * sample_rate_ / 1000.0f);
tomwalters@32:   int max_idx = 0;
tomwalters@32:   float max_val = 0.0f;
tomwalters@32:   for (int i = start_sample; i < buffer_length_; ++i) {
tomwalters@32:     if (sai_temporal_profile[i] > max_val) {
tomwalters@32:       max_idx = i;
tomwalters@32:       max_val = sai_temporal_profile[i];
tomwalters@32:     }
tomwalters@32:   }
tomwalters@32:   return max_idx;
tomwalters@32: }
tomwalters@32: 
tomwalters@12: void ModuleSSI::Process(const SignalBank &input) {
tomwalters@12:   // Check to see if the module has been initialized. If not, processing
tomwalters@12:   // should not continue.
tomwalters@12:   if (!initialized_) {
tomwalters@13:     LOG_ERROR(_T("Module %s not initialized."), module_identifier_.c_str());
tomwalters@12:     return;
tomwalters@12:   }
tomwalters@12: 
tomwalters@12:   // Check that ths input this time is the same as the input passed to
tomwalters@12:   // Initialize()
tomwalters@12:   if (buffer_length_ != input.buffer_length()
tomwalters@12:       || channel_count_ != input.channel_count()) {
tomwalters@12:     LOG_ERROR(_T("Mismatch between input to Initialize() and input to "
tomwalters@13:                  "Process() in module %s."), module_identifier_.c_str());
tomwalters@12:     return;
tomwalters@12:   }
tomwalters@12: 
tomwalters@15:   output_.set_start_time(input.start_time());
tomwalters@12: 
tomwalters@32:   int pitch_index = buffer_length_ - 1;
tomwalters@32:   if (do_pitch_cutoff_) {
tomwalters@32:     pitch_index = ExtractPitchIndex(input);
tomwalters@32:   }
tomwalters@32: 
tomwalters@15:   for (int ch = 0; ch < channel_count_; ++ch) {
tomwalters@32:     float centre_frequency = input.centre_frequency(ch);
tomwalters@15:     // Copy the buffer from input to output, addressing by h-value
tomwalters@15:     for (int i = 0; i < ssi_width_samples_; ++i) {
tomwalters@32:       float h;
tomwalters@32:       float cycle_samples = sample_rate_ / centre_frequency;
tomwalters@32:       if (log_cycles_axis_) {
tomwalters@32:         float gamma_min = -1.0f;
tomwalters@32:         float gamma_max = log2(ssi_width_cycles_);
tomwalters@32:         float gamma = gamma_min + (gamma_max - gamma_min)
tomwalters@32:                                    * static_cast<float>(i)
tomwalters@32:                                    / static_cast<float>(ssi_width_samples_);
tomwalters@32:         h = pow(2.0f, gamma);
tomwalters@32:       } else {
tomwalters@32:         h = static_cast<float>(i) * ssi_width_cycles_
tomwalters@32:             / static_cast<float>(ssi_width_samples_);
tomwalters@32:       }
tomwalters@12: 
tomwalters@15:       // The index into the input array is a floating-point number, which is
tomwalters@15:       // split into a whole part and a fractional part. The whole part and
tomwalters@15:       // fractional part are found, and are used to linearly interpolate
tomwalters@15:       // between input samples to yield an output sample.
tomwalters@15:       double whole_part;
tomwalters@15:       float frac_part = modf(h * cycle_samples, &whole_part);
tomwalters@32:       int sample = floor(whole_part);
tomwalters@32: 
tomwalters@32:       float weight = 1.0f;
tomwalters@32: 
tomwalters@32:       int cutoff_index = buffer_length_ - 1;
tomwalters@32:       if (do_pitch_cutoff_) {
tomwalters@32:         if (pitch_index < cutoff_index) {
tomwalters@32:           if (weight_by_cutoff_) {
tomwalters@32:             weight *= static_cast<float>(buffer_length_)
tomwalters@32:                       / static_cast<float>(pitch_index);
tomwalters@32:           }
tomwalters@32:           cutoff_index = pitch_index;
tomwalters@32:         }
tomwalters@32:       }
tomwalters@32: 
tomwalters@32:       if (weight_by_scaling_) {
tomwalters@32:         if (centre_frequency > pivot_cf_) {
tomwalters@32:           weight *= (centre_frequency / pivot_cf_);
tomwalters@32:         }
tomwalters@32:       }
tomwalters@15: 
tomwalters@15:       float val;
tomwalters@32:       if (sample < cutoff_index) {
tomwalters@15:         float curr_sample = input.sample(ch, sample);
tomwalters@15:         float next_sample = input.sample(ch, sample + 1);
tomwalters@32:         val = weight * (curr_sample
tomwalters@32:                         + frac_part * (next_sample - curr_sample));
tomwalters@15:       } else {
tomwalters@36:         val = 0.0f;
tomwalters@15:       }
tomwalters@15:       output_.set_sample(ch, i, val);
tomwalters@15:     }
tomwalters@15:   }
tomwalters@12:   PushOutput();
tomwalters@12: }
tomwalters@12: }  // namespace aimc
tomwalters@12: