static private List<User> LoadDataset(string datasetFileName)
{
//Get The dataset folder path.
var splittedPath = datasetFileName.Split('\\');
string folderPath = "";
for (int i = 0; i < splittedPath.Length - 1; i++)
{
folderPath += splittedPath[i] + '\\';
}
folderPath += "audiofiles\\";
//read the training samples files names
Dictionary<string, User> users = new Dictionary<string, User>();
StreamReader reader = new StreamReader(datasetFileName);
string line;
while ((line = reader.ReadLine()) != null)
{
string userName = line.Split('/')[0];
string fileName = line.Split('/')[1] + ".wav";
//check if user already exists, if not add an entry in the dictionary.
if (users.ContainsKey(userName) == false)
{
User user = new User();
user.UserTemplates = new List<AudioSignal>();
user.UserName = userName;
users.Add(userName, user);
}
AudioSignal audio;
string fullFileName = folderPath + userName + '\\' + fileName;
try
{
audio = openNISTWav(fullFileName);
}
catch (Exception)
{
audio = AudioOperations.OpenAudioFile(folderPath + userName + '\\' + fileName);
}
audio = AudioOperations.RemoveSilence(audio);
users[userName].UserTemplates.Add(audio);
}
reader.Close();
//move the users to a list for convenience reasons only.
List<User> usersList = new List<User>();
foreach (var user in users)
{
usersList.Add(user.Value);
}
return usersList;
}
static private List<User> ConcatenateSamples(List<User> dataset, int shrinkagefactor)
{
List<User> newDataset = new List<User>();
foreach (User user in dataset)
{
int numberOfSequences = user.UserTemplates.Count;
//NOTE: i didn't handle the case if the number of sequences is not divisible by the shrinkage factor :)
int newNumberOfSequences = numberOfSequences / shrinkagefactor;
User concUser = new User();
concUser.UserName = user.UserName;
concUser.UserTemplates = new List<AudioSignal>(newNumberOfSequences);
int startIndex = 0;
for (int i = 0; i < newNumberOfSequences; i++)
{
int currentConcSeqLength = 0;
double currentConcSeqDuration = 0;
for (int j = startIndex; j < startIndex + shrinkagefactor; j++)
{
currentConcSeqLength += user.UserTemplates[j].data.Length;
currentConcSeqDuration += user.UserTemplates[j].signalLengthInMilliSec;
}
concUser.UserTemplates.Add(new AudioSignal());
concUser.UserTemplates[i].sampleRate = user.UserTemplates[0].sampleRate;
concUser.UserTemplates[i].signalLengthInMilliSec = currentConcSeqDuration;
concUser.UserTemplates[i].data = new double[currentConcSeqLength];
int concIndex = 0;
for (int j = startIndex; j < startIndex + shrinkagefactor; j++)
{
user.UserTemplates[j].data.CopyTo(concUser.UserTemplates[i].data, concIndex);
concIndex += user.UserTemplates[j].data.Length;
}
startIndex += shrinkagefactor;
}
newDataset.Add(concUser);
}
return newDataset;
}
