train_segmenter.cpp File Reference


Classes

struct  Thread_Train_Data_Cluster

Functions

void * thread_train_cluster (void *ptr)

Variables

int SEGMENT_INITIAL_DATA_SEGMENTS = 2
int MINPASSES
int nrOfStateHistPruning
int globalIt = 1
int MERGE_METHOD = 0
int STOP_METHOD = 0
Thread_Train_Data_Cluster trainThreadCluster [MAX_NR_THREADS]
bool cluster_threadRunning [MAX_NR_THREADS]
bool cluster_threadMayStart [MAX_NR_THREADS]
pthread_mutex_t condition_cluster_mutexStart [MAX_NR_THREADS]
pthread_cond_t condition_cluster_threadStart [MAX_NR_THREADS]
pthread_t cluster_thread [MAX_NR_THREADS]
pthread_mutex_t condition_cluster_mutexDone
pthread_cond_t condition_cluster_threadDone
bool loadedOverlapAM = false
FeaturePoolpass2Pool = NULL
int pass2PoolindexWeight = 0
int pass2PoolindexShift = 0

Function Documentation


Variable Documentation

pthread_t cluster_thread[MAX_NR_THREADS]

pthread_mutex_t condition_cluster_mutexStart[MAX_NR_THREADS]

pthread_cond_t condition_cluster_threadStart[MAX_NR_THREADS]

bool loadedOverlapAM = false

Todo:
Docs

Referenced by Train_Segmenter::train().

int MERGE_METHOD = 0

int MINPASSES

The main method of this class. It will 'train' new clusters with maximal maxClusters. If maxClusters is equal to zero, regular speaker clustering is performed. If not, the following steps are added:

  • Merge small clusters (clusters with not enough data used during training) to their closest clusters until all clusters contain are big enough.
  • Merge the remaining clusters until no more than maxClusters clusters are left.

int STOP_METHOD = 0