libjulius/src/confnet.c

Confusion network の生成 [詳細]

#include <julius/julius.h>

ソースコードを見る。

マクロ定義

#define PREFER_GRAPH_CM

Define to enable debug output.

#define BUNDLE_WORD_WITH_SAME_OUTPUT

Julius identify the words by their dictionary IDs, so words with different entries are treated as a different word.

#define m2i(A, B) (B) * order_matrix_count + (A)

Macro to access the order matrix.

関数

static boolean is_same_word (WORD_ID w1, WORD_ID w2, WORD_INFO *winfo)

Determine whether the two words are idential in confusion network generation.

static boolean graph_ordered (int i, int j)

Judge order between two words by their word graph ID.

static void graph_update_order ()

Scan the order matrix to update it at initial step and after word (set) marging.

void graph_make_order (WordGraph *root, RecogProcess *r)

Extract order relationship between any two words in the word graph for confusion network generation.

void graph_free_order ()

Free the order relation data.

static CN_CLUSTER * cn_new ()

Create a new cluster holder.

static void cn_free (CN_CLUSTER *c)

Free a cluster holder

void cn_free_all (CN_CLUSTER **croot)

Free all cluster holders.

static void cn_add_wg (CN_CLUSTER *c, WordGraph *wg)

Add a graph word to a cluster holder.

static void cn_merge (CN_CLUSTER *dst, CN_CLUSTER *src)

Merge a cluster holder into another.

static void cn_destroy (CN_CLUSTER *target, CN_CLUSTER **root)

Erase a cluster holder and remove it from the list.

static void cn_build_wordlist (CN_CLUSTER *c, WORD_INFO *winfo)

Build / update word list from graph words for a cluster holder.

static int compare_cluster (CN_CLUSTER **x, CN_CLUSTER **y)

qsort callback to sort clusters by their time order.

static PROB get_intraword_similarity (WordGraph *w1, WordGraph *w2)

Compute intra-word similarity of two graph words for confusion network generation.

static PROB get_cluster_intraword_similarity (CN_CLUSTER *c1, CN_CLUSTER *c2, WORD_INFO *winfo)

Compute intra-word similarity of two clusters.

static int minimum (int a, int b, int c)

Return minimum value of the three arguments.

static int edit_distance (WORD_ID w1, WORD_ID w2, WORD_INFO *winfo)

Calculate Levenstein distance (edit distance) of two words.

static PROB get_cluster_interword_similarity (CN_CLUSTER *c1, CN_CLUSTER *c2, WORD_INFO *winfo)

Compute inter-word similarity of two clusters.

CN_CLUSTER * confnet_create (WordGraph *root, RecogProcess *r)

Create a confusion network from word graph.

変数

static char * order_matrix = NULL

Temporal matrix work area to hold the order relations between words.

static int order_matrix_count

Number of words to be expressed in the order matrix.

説明

Confusion network の生成

認識の結果得られた単語グラフから，confusion network を生成する.

作者:: Akinobu Lee

日付:: Thu Aug 16 00:15:51 2007

Revision: 1.1.1.1

confnet.c で定義されています。

マクロ定義

#define PREFER_GRAPH_CM

Define to enable debug output.

Define to enable further debug output. Use graph-based CM for confusion network generation. If not defined search-based CM (default of old julius) will be used. However, the clustering process does not work properly with this definition, since sum of the search- based CM for a word set on the same position is not always 1.0. Thus you'd better always define this.

confnet.c の 52 行で定義されています。

#define BUNDLE_WORD_WITH_SAME_OUTPUT

Julius identify the words by their dictionary IDs, so words with different entries are treated as a different word.

If this is defined, Julius treat words with the same output string as same words and bundle them in confusion network generation.

confnet.c の 61 行で定義されています。

参照元 is_same_word().

関数

static boolean is_same_word	(	WORD_ID	w1,
		WORD_ID	w2,
		WORD_INFO *	winfo
	)			`[static]`

Determine whether the two words are idential in confusion network generation.

引数:

	w1	[in] first word
	w2	[in] second word
	winfo	[in] word dictionary

戻り値:: TRUE if they are idential, FALSE if not.

confnet.c の 75 行で定義されています。

参照元 cn_build_wordlist(), と get_cluster_intraword_similarity().

static boolean graph_ordered	(	int	i,
		int	j
	)			`[static]`

Judge order between two words by their word graph ID.

引数:

	i	[in] id of left graph word
	j	[in] id of right graph word

戻り値:: TRUE if they are ordered, or FALSE if not.

confnet.c の 114 行で定義されています。

void graph_make_order	(	WordGraph *	root,
		RecogProcess *	r
	)

Extract order relationship between any two words in the word graph for confusion network generation.

引数:

	root	[in] root pointer to the word graph
	r	[in] recognition process instance

confnet.c の 166 行で定義されています。

static CN_CLUSTER* cn_new ( ) [static]

Create a new cluster holder.

戻り値:: the newly allocated cluster holder.

confnet.c の 229 行で定義されています。

static void cn_free ( CN_CLUSTER * c ) [static]

Free a cluster holder

引数:

[out] a cluster holder to be released.

confnet.c の 249 行で定義されています。

参照元 cn_destroy(), と cn_free_all().

void cn_free_all ( CN_CLUSTER ** croot )

Free all cluster holders.

引数:

croot

[out] pointer to root pointer of cluster holder list.

confnet.c の 267 行で定義されています。

参照元 clear_result().

呼出しグラフ:

static void cn_add_wg	(	CN_CLUSTER *	c,
		WordGraph *	wg
	)			`[static]`

Add a graph word to a cluster holder.

引数:

	c	[out] cluster holder
	wg	[in] graph word to be added

confnet.c の 286 行で定義されています。

参照元 cn_merge().

static void cn_merge	(	CN_CLUSTER *	dst,
		CN_CLUSTER *	src
	)			`[static]`

Merge a cluster holder into another.

引数:

	dst	[i/o] target cluster holder
	src	[in] source cluster holder.

confnet.c の 303 行で定義されています。

static void cn_destroy	(	CN_CLUSTER *	target,
		CN_CLUSTER **	root
	)			`[static]`

Erase a cluster holder and remove it from the list.

引数:

	target	[i/o] a cluster holder to be erased
	root	[i/o] pointer to root pointer of cluster holder list

confnet.c の 334 行で定義されています。

static void cn_build_wordlist	(	CN_CLUSTER *	c,
		WORD_INFO *	winfo
	)			`[static]`

Build / update word list from graph words for a cluster holder.

引数:

	c	[i/o] cluster holder to process
	winfo	[in] word dictionary

confnet.c の 360 行で定義されています。

static int compare_cluster	(	CN_CLUSTER **	x,
		CN_CLUSTER **	y
	)			`[static]`

qsort callback to sort clusters by their time order.

引数:

	x	[in] element 1
	y	[in] element 2

戻り値:: order value

confnet.c の 389 行で定義されています。

static PROB get_intraword_similarity	(	WordGraph *	w1,
		WordGraph *	w2
	)			`[static]`

Compute intra-word similarity of two graph words for confusion network generation.

引数:

	w1	[in] graph word 1
	w2	[in] graph word 2

戻り値:: the similarity value.

confnet.c の 428 行で定義されています。

参照元 get_cluster_intraword_similarity().

static PROB get_cluster_intraword_similarity	(	CN_CLUSTER *	c1,
		CN_CLUSTER *	c2,
		WORD_INFO *	winfo
	)			`[static]`

Compute intra-word similarity of two clusters.

引数:

	c1	[in] cluster 1
	c2	[in] cluster 2
	winfo	[in] word dictionary

戻り値:: the maximum similarity.

confnet.c の 488 行で定義されています。

static int minimum	(	int	a,
		int	b,
		int	c
	)			`[static]`

Return minimum value of the three arguments.

引数:

	a	[in] value 1
	b	[in] value 2
	c	[in] value 3

戻り値:: the minumum value.

confnet.c の 535 行で定義されています。

static int edit_distance	(	WORD_ID	w1,
		WORD_ID	w2,
		WORD_INFO *	winfo
	)			`[static]`

Calculate Levenstein distance (edit distance) of two words.

引数:

	w1	[in] word ID 1
	w2	[in] word ID 2
	winfo	[in] word dictionary

戻り値:: the distance.

confnet.c の 557 行で定義されています。

static PROB get_cluster_interword_similarity	(	CN_CLUSTER *	c1,
		CN_CLUSTER *	c2,
		WORD_INFO *	winfo
	)			`[static]`

Compute inter-word similarity of two clusters.

引数:

	c1	[in] cluster 1
	c2	[in] cluster 2
	winfo	[in] word dictionary

戻り値:: the average similarity.

confnet.c の 603 行で定義されています。

CN_CLUSTER* confnet_create	(	WordGraph *	root,
		RecogProcess *	r
	)

Create a confusion network from word graph.

引数:

	root	[in] root pointer of word graph
	r	[in] recognition process instance

戻り値:: root pointer to the cluster list.

confnet.c の 702 行で定義されています。

Juliusに対してTue Dec 18 16:04:49 2007に生成されました。

1.5.4


マクロ定義
#define	PREFER_GRAPH_CM
	Define to enable debug output.
#define	BUNDLE_WORD_WITH_SAME_OUTPUT
	Julius identify the words by their dictionary IDs, so words with different entries are treated as a different word.
#define	m2i(A, B) (B) * order_matrix_count + (A)
	Macro to access the order matrix.
関数
static boolean	is_same_word (WORD_ID w1, WORD_ID w2, WORD_INFO *winfo)
	Determine whether the two words are idential in confusion network generation.
static boolean	graph_ordered (int i, int j)
	Judge order between two words by their word graph ID.
static void	graph_update_order ()
	Scan the order matrix to update it at initial step and after word (set) marging.
void	graph_make_order (WordGraph root, RecogProcess r)
	Extract order relationship between any two words in the word graph for confusion network generation.
void	graph_free_order ()
	Free the order relation data.
static CN_CLUSTER *	cn_new ()
	Create a new cluster holder.
static void	cn_free (CN_CLUSTER *c)
	Free a cluster holder
void	cn_free_all (CN_CLUSTER **croot)
	Free all cluster holders.
static void	cn_add_wg (CN_CLUSTER c, WordGraph wg)
	Add a graph word to a cluster holder.
static void	cn_merge (CN_CLUSTER dst, CN_CLUSTER src)
	Merge a cluster holder into another.
static void	cn_destroy (CN_CLUSTER target, CN_CLUSTER *root)
	Erase a cluster holder and remove it from the list.
static void	cn_build_wordlist (CN_CLUSTER c, WORD_INFO winfo)
	Build / update word list from graph words for a cluster holder.
static int	compare_cluster (CN_CLUSTER x, CN_CLUSTER y)
	qsort callback to sort clusters by their time order.
static PROB	get_intraword_similarity (WordGraph w1, WordGraph w2)
	Compute intra-word similarity of two graph words for confusion network generation.
static PROB	get_cluster_intraword_similarity (CN_CLUSTER c1, CN_CLUSTER c2, WORD_INFO *winfo)
	Compute intra-word similarity of two clusters.
static int	minimum (int a, int b, int c)
	Return minimum value of the three arguments.
static int	edit_distance (WORD_ID w1, WORD_ID w2, WORD_INFO *winfo)
	Calculate Levenstein distance (edit distance) of two words.
static PROB	get_cluster_interword_similarity (CN_CLUSTER c1, CN_CLUSTER c2, WORD_INFO *winfo)
	Compute inter-word similarity of two clusters.
CN_CLUSTER *	confnet_create (WordGraph root, RecogProcess r)
	Create a confusion network from word graph.
変数
static char *	order_matrix = NULL
	Temporal matrix work area to hold the order relations between words.
static int	order_matrix_count
	Number of words to be expressed in the order matrix.