ronw@642: // Copyright 2013, Google, Inc.
ronw@642: // Author: Ron Weiss <ronw@google.com>
ronw@642: //
ronw@642: // This C++ file is part of an implementation of Lyon's cochlear model:
ronw@642: // "Cascade of Asymmetric Resonators with Fast-Acting Compression"
ronw@642: // to supplement Lyon's upcoming book "Human and Machine Hearing"
ronw@642: //
ronw@642: // Licensed under the Apache License, Version 2.0 (the "License");
ronw@642: // you may not use this file except in compliance with the License.
ronw@642: // You may obtain a copy of the License at
ronw@642: //
ronw@642: //     http://www.apache.org/licenses/LICENSE-2.0
ronw@642: //
ronw@642: // Unless required by applicable law or agreed to in writing, software
ronw@642: // distributed under the License is distributed on an "AS IS" BASIS,
ronw@642: // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
ronw@642: // See the License for the specific language governing permissions and
ronw@642: // limitations under the License.
ronw@642: 
ronw@642: #ifndef CARFAC_SAI_H_
ronw@642: #define CARFAC_SAI_H_
ronw@642: 
ronw@642: #include <vector>
ronw@642: 
alexbrandmeyer@643: #include "common.h"
ronw@642: 
ronw@642: // Design parameters for a single SAI.
ronw@642: struct SAIParams {
ronw@642:   // Number of channels (height) of the SAI.
ronw@654:   int num_channels;
ronw@642: 
ronw@642:   // TODO(ronw): Consider parameterizing this as past_lags and
ronw@642:   // future_lags, with width == past_lags + 1 + future_lags.
ronw@642:   //
ronw@642:   // Total width (i.e. number of lag samples) of the SAI.
ronw@642:   int width;
ronw@642:   // Number of lag samples that should come from the future.
ronw@642:   int future_lags;
ronw@642:   // Number of windows (triggers) to consider during each SAI frame.
ronw@654:   int num_window_pos;
ronw@642: 
ronw@642:   // TODO(ronw): more carefully define terms "window" and "frame"
ronw@642: 
ronw@642:   // Size of the window to compute.
ronw@642:   int window_width;
ronw@642: 
ronw@642:   FPType channel_smoothing_scale;
ronw@642: };
ronw@642: 
ronw@642: class SAI {
ronw@642:  public:
ronw@642:   explicit SAI(const SAIParams& params);
ronw@642: 
ronw@650:   // Reinitialize using the specified parameters.
ronw@650:   void Redesign(const SAIParams& params);
ronw@650: 
ronw@650:   // Reset the internal state.
ronw@650:   void Reset();
ronw@650: 
ronw@654:   // Fills output_frame with a params_.num_channels by params_.width SAI frame
ronw@642:   // computed from the given input frames.
ronw@642:   //
ronw@642:   // The input should have dimensionality of params_.window_width by
ronw@654:   // params_.num_channels.  Inputs containing too few frames are zero-padded.
alexbrandmeyer@643:   // FIXME: ArrayXX input type would be less awkward.
alexbrandmeyer@643:   void RunSegment(const std::vector<ArrayX>& input,
alexbrandmeyer@643:                   ArrayXX* output_output_frame);
ronw@642: 
ronw@642:  private:
ronw@649:   // Processes successive windows within input_buffer, choose trigger
ronw@642:   // points, and blend each window into output_buffer.
alexbrandmeyer@643:   void StabilizeSegment(const ArrayXX& input_buffer,
alexbrandmeyer@643:                         ArrayXX* output_buffer) const;
ronw@642: 
ronw@642:   SAIParams params_;
ronw@642:   // Window function to apply before selecting a trigger point.
ronw@642:   // Size: params_.window_width.
alexbrandmeyer@643:   ArrayX window_;
ronw@642:   // Buffer to store a large enough window of input frames to compute
ronw@654:   // a full SAI frame.  Size: params_.num_channels by params_.buffer_width.
alexbrandmeyer@643:   ArrayXX input_buffer_;
ronw@654:   // Output frame buffer.  Size: params_.num_channels by params_.width.
alexbrandmeyer@643:   ArrayXX output_buffer_;
ronw@642: };
ronw@642: 
ronw@646: #endif  // CARFAC_SAI_H_