gyorgyf@60: /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */
gyorgyf@60: 
gyorgyf@60: /*
gyorgyf@60:     Sonic Annotator
gyorgyf@60:     A utility for batch feature extraction from audio files.
gyorgyf@60:     Mark Levy, Chris Sutton and Chris Cannam, Queen Mary, University of London.
gyorgyf@60:     Copyright 2007-2008 QMUL.
gyorgyf@60: 
gyorgyf@60:     This program is free software; you can redistribute it and/or
gyorgyf@60:     modify it under the terms of the GNU General Public License as
gyorgyf@60:     published by the Free Software Foundation; either version 2 of the
gyorgyf@60:     License, or (at your option) any later version.  See the file
gyorgyf@60:     COPYING included with this distribution for more information.
gyorgyf@60: */
gyorgyf@60: 
gyorgyf@60: #include <fstream>
gyorgyf@60: 
gyorgyf@60: #include <QFileInfo>
gyorgyf@60: 
gyorgyf@60: #include <ctime>
gyorgyf@60: 
gyorgyf@60: #include <stdint.h>
gyorgyf@60: 
gyorgyf@60: #include "BinaryFeatureWriter.h"
gyorgyf@60: 
gyorgyf@60: #include "base/RealTime.h"
gyorgyf@60: 
gyorgyf@60: #include "../version.h"
gyorgyf@60: 
gyorgyf@60: #ifdef _WIN32
gyorgyf@60: #define platform "Windows"
gyorgyf@60: #elif __APPLE__
gyorgyf@60: #define platform "MacOS"
gyorgyf@60: #elif __linux__
gyorgyf@60: #define platform "Linux"
gyorgyf@60: #else
gyorgyf@60: #define platform "Unix"
gyorgyf@60: #endif
gyorgyf@60: #ifdef __LP64__ //__x86_64__
gyorgyf@60: #define arch "64" 
gyorgyf@60: #elif _WIN64
gyorgyf@60: #define arch "64" 
gyorgyf@60: #else
gyorgyf@60: #define arch "32"
gyorgyf@60: #endif
gyorgyf@60: 
gyorgyf@60: # define MAJOR_VERSION_PY 0
gyorgyf@60: 
gyorgyf@60: using namespace std;
gyorgyf@60: using namespace Vamp;
gyorgyf@60: 
gyorgyf@60: // Parameter names
gyorgyf@60: string
gyorgyf@60: BinaryFeatureWriter::outputFileParam = "output";
gyorgyf@60: 
gyorgyf@60: struct BinaryFeatureWriter::OutputStream
gyorgyf@60: {
gyorgyf@60:     ofstream* stream;
gyorgyf@60: 	bool newtransform;
gyorgyf@60: 	const Transform *transform;
gyorgyf@60: 	OutputStream() : newtransform(true),stream(NULL),transform(NULL) { }
gyorgyf@60: 	~OutputStream() { if (stream != NULL) {stream->close(); delete stream;} }
gyorgyf@60: 	
gyorgyf@60: 	struct header_t { 
gyorgyf@66: 		int16_t BOM16;          // 16-bit BOM (FEFF as in UTF-16) 2 bytes
gyorgyf@66: 		int32_t BOM32;          // 32-bit BOM (human readable: e.g. ABCD) 4 bytes
gyorgyf@64: 		char major_version;     // check for binary compatibility
gyorgyf@64: 		char minor_version;     // changes in txt parts only
gyorgyf@64: 		char compression;       // use of stream compression (e.g. gzip)
gyorgyf@64: 		char reserved1;         // reserved byte
gyorgyf@64: 		char reserved2;         // reserved byte
gyorgyf@64: 		char float_size;        // size of float
gyorgyf@64: 		char int_size;          // size of int
gyorgyf@64: 		char info[160];         // 160 byte text field
gyorgyf@64: 		char null; };           // NULL
gyorgyf@60: 
gyorgyf@60: 	bool open(string filename, bool append = true) {
gyorgyf@60: 
gyorgyf@60: 		if (stream) return true;
gyorgyf@60: 		
gyorgyf@66: 		header_t header = {(int16_t) 0xFEFF, (int32_t) 0x41424344,MAJOR_VERSION_PY,2,0,NULL,NULL,(char)sizeof(float),(char)sizeof(int),{NULL},NULL}; 	
gyorgyf@60: 		char* p_header = reinterpret_cast<char*>(&header);
gyorgyf@60: 
gyorgyf@60: 		if (append) 
gyorgyf@60: 			stream = new ofstream(filename.c_str(), fstream::binary | ios_base::out | ios_base::in | ofstream::ate);
gyorgyf@60: 		else 
gyorgyf@60: 			stream = new ofstream(filename.c_str(), fstream::binary);
gyorgyf@65: 		
gyorgyf@65: 		if (!stream || !stream->is_open()) {
gyorgyf@65: 			if (append) {
gyorgyf@65: 				cerr << endl << "NOTE: Writing new binary output file: " << filename << endl;
gyorgyf@65: 				delete stream; 
gyorgyf@65: 				stream = NULL;
gyorgyf@65: 				return open(filename,false);
gyorgyf@65: 			} else {
gyorgyf@65: 				cerr << "ERROR: BinaryFeatureWriter::OutputStream::open(): can't open file " << filename << endl;
gyorgyf@65: 		        return false;
gyorgyf@65: 			}
gyorgyf@60: 		}
gyorgyf@60: 		
gyorgyf@60: 		// verify input file format
gyorgyf@60: 		if (append) {
gyorgyf@60: 			ifstream istream;
gyorgyf@60: 			istream.open(filename.c_str(), fstream::binary | ios::in);
gyorgyf@60: 			if (!istream.is_open()) { 
gyorgyf@60: 				cerr << "ERROR: BinaryFeatureWriter::OutputStream::open(): Can not verify supplied output stream." << endl;
gyorgyf@60: 				return false;
gyorgyf@60: 			}
gyorgyf@60: 			header_t iheader;
gyorgyf@60: 			istream.read(reinterpret_cast<char*>(&iheader),sizeof(header_t));
gyorgyf@60: 			istream.close();
gyorgyf@60: 			int16_t FEFF = 0xFEFF;
gyorgyf@60: 			if (iheader.BOM16 != FEFF) {
gyorgyf@60: 				if (iheader.BOM16 == (int16_t) 0xFFFE) {
gyorgyf@60: 					cerr << "ERROR: BinaryFeatureWriter::OutputStream::open(): This file apperas to have created on a different platform. Can not be appended. " 
gyorgyf@60: 					<< "Byte order mark: " << iheader.BOM32 << endl;
gyorgyf@60: 					return false;
gyorgyf@60: 				} else {
gyorgyf@60: 					cerr << "ERROR: BinaryFeatureWriter::OutputStream::open(): Invalid target file for this writer." << endl;
gyorgyf@60: 					return false;
gyorgyf@60: 				}
gyorgyf@60: 			}
gyorgyf@60: 			if (iheader.major_version != (char) MAJOR_VERSION_PY || iheader.BOM32 != (int32_t) 0x41424344) {
gyorgyf@60: 				cerr << "ERROR: BinaryFeatureWriter::OutputStream::open(): This file is not binary compatible with this version of the writer." 
gyorgyf@60: 				<< "file version: " << iheader.major_version << " required: " << MAJOR_VERSION_PY << endl;
gyorgyf@60: 				return false;
gyorgyf@60: 			}
gyorgyf@60: 		}
gyorgyf@60: 
gyorgyf@60: 		if (append) stream->seekp(0,ios_base::end);	
gyorgyf@60: 
gyorgyf@60: 		if (!append) { 
gyorgyf@60: 			time_t now = time(0);
gyorgyf@60: 			tm* gmtm = gmtime(&now);
gyorgyf@60: 			QString timestamp;
gyorgyf@60: 			if (gmtm != NULL) 
gyorgyf@60: 				timestamp = QString("%1 %2").arg(", Created: ").arg(asctime(gmtm)).trimmed();
gyorgyf@66: 			string info = QString("_SONIC ANNOTATOR v%1 PYTHON BINARY V0.2, Platform: %2-%3bit%4")
gyorgyf@60: 				.arg(RUNNER_VERSION)
gyorgyf@60: 				.arg(platform)
gyorgyf@60: 				.arg(arch)
gyorgyf@60: 				.arg(timestamp).toStdString();
gyorgyf@60: 			strncpy(reinterpret_cast<char*>(&header.info),info.c_str(), info.length() <= 160 ? info.length() : 160);
gyorgyf@60: 			stream->write(p_header,sizeof(header_t));
gyorgyf@60: 		}
gyorgyf@60: 		return true;
gyorgyf@60: 	}    
gyorgyf@60: };
gyorgyf@60: 
gyorgyf@60: 
gyorgyf@60: BinaryFeatureWriter::BinaryFeatureWriter() : 
gyorgyf@60:     outputFile("features")
gyorgyf@60: {
gyorgyf@60: 	binary = new OutputStream();
gyorgyf@60: }
gyorgyf@60: 
gyorgyf@60: BinaryFeatureWriter::~BinaryFeatureWriter()
gyorgyf@60: {
gyorgyf@60: 	if (binary) delete binary;
gyorgyf@60: }
gyorgyf@60: 
gyorgyf@60: BinaryFeatureWriter::ParameterList
gyorgyf@60: BinaryFeatureWriter::getSupportedParameters() const
gyorgyf@60: {
gyorgyf@60:     ParameterList pl;
gyorgyf@60:     Parameter p;
gyorgyf@60: 
gyorgyf@60:     p.name = outputFileParam;
gyorgyf@60:     p.description = "Binary output file path";
gyorgyf@60:     p.hasArg = true;
gyorgyf@60:     pl.push_back(p);
gyorgyf@60: 
gyorgyf@60:     return pl;
gyorgyf@60: }
gyorgyf@60: 
gyorgyf@60: void
gyorgyf@60: BinaryFeatureWriter::setParameters(map<string, string> &params)
gyorgyf@60: {
gyorgyf@60:     if (params.find(outputFileParam) != params.end()) {
gyorgyf@60:         setOutputFile(params[outputFileParam]);
gyorgyf@60:         params.erase(outputFileParam);
gyorgyf@60:     }
gyorgyf@60: }
gyorgyf@60: 
gyorgyf@60: void
gyorgyf@60: BinaryFeatureWriter::setOutputFile(const string &file)
gyorgyf@60: {
gyorgyf@60:     outputFile = file;
gyorgyf@60: }
gyorgyf@60: 
gyorgyf@60: void BinaryFeatureWriter::write(QString trackid,
gyorgyf@60:                                  const Transform &transform,
gyorgyf@60:                                  const Vamp::Plugin::OutputDescriptor& output, 
gyorgyf@60:                                  const Vamp::Plugin::FeatureList& featureList,
gyorgyf@60:                                  std::string summaryType)
gyorgyf@60: {
gyorgyf@60:     //!!! use summaryType
gyorgyf@60:     if (summaryType != "") {
gyorgyf@60:         //!!! IMPLEMENT
gyorgyf@60:         cerr << "ERROR: BinaryFeatureWriter::write: Writing summaries is not yet implemented!" << endl;
gyorgyf@60:         exit(1);
gyorgyf@60:     }
gyorgyf@60: 
gyorgyf@60: 	// TODO: Consider writing out NumPy arrays directly following this documentation:
gyorgyf@60: 	// https://github.com/numpy/numpy/blob/master/doc/neps/npy-format.txt 
gyorgyf@60: 	// and using the .npy format
gyorgyf@60: 		
gyorgyf@60: 	// return if file could not be opened
gyorgyf@60: 	if(!openBinaryFile()) {
gyorgyf@60: 		cerr << "ERROR: BinaryFeatureWriter::write: Error opening binary output file!" << endl;
gyorgyf@60: 	    exit(1);	        
gyorgyf@60: 	}	
gyorgyf@60: 		
gyorgyf@60: 	ofstream &ofs = *(binary->stream); 
gyorgyf@60: 	
gyorgyf@60: 	// The manager does not call finish() after writing different outputs from the same plugin, but we need this behaviour here:
gyorgyf@60: 	if (!binary->newtransform && binary->transform != NULL && binary->transform != &transform) finish();
gyorgyf@60: 	
gyorgyf@60: 	// write a python dictionary string containing (some) metadata needed to interpret the results
gyorgyf@60: 	// this can be evaluated in python using the expression : d = eval(f.readline())
gyorgyf@60: 	// given f is an open file, which should yield a valid dictionary. 
gyorgyf@60: /*	
gyorgyf@60: 	enum SampleType {
gyorgyf@60: 	
gyorgyf@60: 	    /// Results from each process() align with that call's block start
gyorgyf@60: 	    0: OneSamplePerStep, 
gyorgyf@60: 	
gyorgyf@60: 	    /// Results are evenly spaced in time 
gyorgyf@60: 	    1: FixedSampleRate,
gyorgyf@60: 	
gyorgyf@60: 	    /// Results are unevenly spaced and have individual timestamps
gyorgyf@60: 	    2: VariableSampleRate 
gyorgyf@60: 	};
gyorgyf@60: */
gyorgyf@60: 	
gyorgyf@60: 	if (binary->newtransform) {
gyorgyf@60: 		binary->newtransform = false;
gyorgyf@60: 		output_binCount = output.binCount;
gyorgyf@69: 		feature_count = 0;
gyorgyf@69: 		element_count = 0;
gyorgyf@60: 		binary->transform = &transform;
gyorgyf@60: 	
gyorgyf@60: 	ofs << endl << "{"
gyorgyf@60: 	<< "\"track_id\":\"" << trackid << "\"," 
gyorgyf@60: 	<< "\"transform_id\":\"" << transform.getIdentifier() << "\","
gyorgyf@60: 	<< "\"sample_rate\":" << transform.getSampleRate() << ","
gyorgyf@60: 	<< "\"step_size\":" << transform.getStepSize() << ","
gyorgyf@60: 	<< "\"block_size\":" << transform.getBlockSize() << ","
gyorgyf@60: 	<< "\"window_type\":" << transform.getWindowType() << ","
gyorgyf@60: 	
gyorgyf@60: 	<< "\"features_list\":" << featureList.size() << ","
gyorgyf@60: 	<< "\"bin_count\":" << output.binCount << ","
gyorgyf@60: 	// << "\"output_description\":\"" << output.description << "\","
gyorgyf@60: 	<< "\"output_sample_type\":" << output.sampleType << ","
gyorgyf@60: 	<< "\"output_sample_rate\":" << output.sampleRate << ",";
gyorgyf@60: 	
gyorgyf@60: 	// Write start time and duration if the transform is not for the whole file
gyorgyf@60: 	if (transform.getDuration().toString() != "0.000000000") {
gyorgyf@60: 		ofs << "\"start_time\":\"" << transform.getStartTime().toString() << "\",";	
gyorgyf@60: 		ofs << "\"duration\":\"" << transform.getDuration().toString() << "\",";
gyorgyf@60: 	}
gyorgyf@60: 	// Write plugin version if known. (NOTE: using RDF transforms, it remains empty for some reason)
gyorgyf@60: 	if (!transform.getPluginVersion().isEmpty())
gyorgyf@60: 		ofs << "\"plugin_version\":\"" << transform.getPluginVersion() << "\",";
gyorgyf@60: 	
gyorgyf@60: 	// write transform parameters into a dict: parameters:{"parameter_name":value,...} where value is float	
gyorgyf@60: 	ofs << "\"parameters\":{" ;
gyorgyf@60: 	ParameterMap m = transform.getParameters();	
gyorgyf@60: 	
gyorgyf@60: 	for (ParameterMap::const_iterator i = m.begin(); i != m.end(); ++i) 
gyorgyf@60: 		// note last comma is ignored by python
gyorgyf@60: 		if (i == m.begin())
gyorgyf@60: 			ofs << QString("\"%1\":%2").arg(i->first).arg(i->second);
gyorgyf@60: 		else
gyorgyf@60: 			ofs << QString(",\"%1\":%2").arg(i->first).arg(i->second);
gyorgyf@60: 	ofs << "}"; 
gyorgyf@60: 	
gyorgyf@60: 	// write the data size last, and close the line.
gyorgyf@60: 	data_size_pos = ofs.tellp();	
gyorgyf@60: 	ofs << "                                                   }" << endl;	
gyorgyf@60: 	}
gyorgyf@60: 	
gyorgyf@68: 	// BUG: File corruption after about 34000 runs due to beat spectrum plugin returning less data than advertised !!!
gyorgyf@68: 	
gyorgyf@60: 	// write the feature data
gyorgyf@60: 	feature_count += featureList.size();	
gyorgyf@60: 	for (size_t i = 0; i < featureList.size(); ++i) {
gyorgyf@69: 		for (size_t j = 0; j < featureList[i].values.size(); ++j) {
gyorgyf@60: 			ofs.write( (const char*) &featureList[i].values[j], sizeof(featureList[i].values[j]) );
gyorgyf@69: 			element_count++;
gyorgyf@69: 		}
gyorgyf@60: 	}
gyorgyf@60: 	// ofs << endl;
gyorgyf@60: 	
gyorgyf@60: 	// // write time stamp data
gyorgyf@60: 	// for (int i = 0; i < featureList.size(); ++i) {
gyorgyf@60: 	// 	for (int j = 0; j < featureList[i].values.size(); ++j) {
gyorgyf@60: 	// 		// float sec = (int) featureList[i].timestamp.sec;
gyorgyf@60: 	// 		// float nsec = (int) featureList[i].timestamp.nsec;
gyorgyf@60: 	// 		// (*dbfiles[id].ofs).write( (const char*) &sec, sizeof(int));
gyorgyf@60: 	// 		// (*dbfiles[id].ofs).write( (const char*) &nsec, sizeof(int));
gyorgyf@60: 	// 		ofs.write( (const char*) &featureList[i].timestamp.sec, sizeof(int));
gyorgyf@60: 	// 		ofs.write( (const char*) &featureList[i].timestamp.nsec, sizeof(int));
gyorgyf@60: 	// 		
gyorgyf@60: 	// 	}
gyorgyf@60: 	// 		
gyorgyf@60: 	// }
gyorgyf@60: 	
gyorgyf@60: 	// -- UNCOMMENT - TO - HERE --
gyorgyf@60: 		
gyorgyf@60: 	
gyorgyf@60: 
gyorgyf@60: }
gyorgyf@60: 
gyorgyf@60: bool BinaryFeatureWriter::openBinaryFile()
gyorgyf@60: {
gyorgyf@64: 	if (outputFile.rfind(".") == string::npos)
gyorgyf@64: 		return binary->open(outputFile + ".bin");
gyorgyf@64: 	else 
gyorgyf@64: 		return binary->open(outputFile);
gyorgyf@60: }
gyorgyf@60: 
gyorgyf@60: void BinaryFeatureWriter::finish()
gyorgyf@60: {
gyorgyf@60: 	ofstream &ofs = *(binary->stream);
gyorgyf@60: 	binary->newtransform = true;
gyorgyf@60: 	// ofs << endl;
gyorgyf@60: 	long t = ofs.tellp();
gyorgyf@60: 	ofs.seekp(data_size_pos);
gyorgyf@60: 	// fill in the missing information in the transform python dict that is required to read the output into an array
gyorgyf@60: 	ofs << ",\"feature_count\":" 
gyorgyf@60: 		<< feature_count
gyorgyf@60: 		<< ",\"data_size\":" 	
gyorgyf@69: 		// << feature_count * output_binCount * sizeof(float);	
gyorgyf@69: 		<< element_count * sizeof(float);
gyorgyf@69: 	ofs.seekp(t);
gyorgyf@69: 	if (element_count != feature_count * output_binCount) 
gyorgyf@69: 		cerr << "ERROR: The number of feature values written is different from the expected value given by the output bin count and the number of features returned by the plugin. \n"
gyorgyf@69: 		<< "Otput bin count: " << output_binCount << ", Feature count: " << feature_count << ", Expected elements: " << feature_count * output_binCount 
gyorgyf@69: 		<< ", Written elements: " << element_count << ", Resulting float32 data size: " << element_count * sizeof(float) << " bytes."
gyorgyf@69: 		<< endl;
gyorgyf@60:     // FileFeatureWriter::finish();
gyorgyf@60: }