tblLVST_Transcripts

Table Description:

Stores all of the transcriptions for an interaction.

Table structure:

Column Name	Data Type	Data Length	Default Value	See Also
InteractionID	INTEGER	N/A	N/A	N/A
Transcript	VARCHAR	900	N/A	N/A
cTranscriptType	INTEGER	N/A	N/A	tblLVST_Transcripts _cTranscriptType
Gender	CHAR	1	N/A	N/A
Notes	VARCHAR	900	N/A	N/A
DateEntered	TIMESTAMP	N/A	0	N/A
AsideF	Integer	N/A	0	N/A
CheckF	Integer	N/A	0	N/A
ChoppedF	Integer	N/A	0	N/A
PartialF	Integer	N/A	0	N/A
stTranscriber	Integer	N/A	0	N/A
Version	Integer	N/A	0	N/A

Primary Keys:

InteractionID
cTranscriptType
Version

Foreign Key	Reference Table	Reference Key
InteractionID	tblLVST_CallInteractions	InteractionID
cTranscriptType	tblLVST_Transcripts_cTranscriptType	ConstantValue
stTranscriber	tblLVST_StringTable	StringID

Column Descriptions and Constraints:

Column	Description	Constraint Name	Constraint Description
InteractionID	The transcript itself. This can be empty, although that is rare; usually even when the speaker doesn't say anything, there will be noise. The field is searchable, but any searches on are full text searches, and therefore expensive.	N/A	N/A
Transcript	The transcript itself. This can be empty, although that is rare; usually even when the speaker doesn't say anything, there will be noise. The field is searchable, but any searches on are full text searches, and therefore expensive.	N/A	N/A
cTranscriptType	Foreign key to tblLVST_Transcripts_cTranscriptType. Part of the primary key set. Indicates the kind of transcription. Currently, transcripts are either word-for-word ('RAW'), or the semantic result ('SEMANTIC').	TranscriptTypeMustExist	N/A
Gender	Usually, this is the *perceived* sex of the speaker, based on the transcriber's intuition from listening to the audio. It's better if this can be exactly correct, although automated methods are good, too (for example, based on pitch detection).	N/A	N/A
Notes	Notes on the transcription. Searchable, but not indexed, so effectively any search is a full text search of every notes entry.	N/A	N/A
DateEntered	The date this transcription was written. Note that nearly *any* change to the transcription will generate a new entry, including changing the notes, any of the flags, or the transcription itself.	N/A	N/A
AsideF	Marks this transcript as containing speech from the speaker, but not directed at the speech appliction, i.e., the speaker turns their head from the handset, and speaks to someone in the room. Typically, this utterance has to be rejected from usage in training new acoustic models, because the aside speech is really just noise. 1 is true, 0 false.	N/A	N/A
CheckF	Marks this transcript as requiring further review by someone else (usually the transcription manager). LumenVox uses this in cases where a situation arises that isn't in the transcription protocol. 1 is true, 0 false.	N/A	N/A
ChoppedF	Marks this transcript as containing a chopped word, i.e., the end pointing/speech detector started/ended recording inside a word. Typically, this utterance has to be rejected from usage in training new acoustic models, because there won't be a pronunciation for the chopped word in the dictionary, and because there are odd transient effects in the audio. 1 is true, 0 false.	N/A	N/A
PartialF	Marks this transcript as containing a partial word, i.e., the user said a word which wasn't completed. Typically, this utterance has to be rejected from usage in training new acoustic models, because there won't be a pronunciation for the partial word in the dictionary. 1 is true, 0 false.	N/A	N/A
stTranscriber	Foreign key to tblLVST_StringTable. Identifies the transcriber for this version of this transcript.	N/A	N/A
Version	Identifies the transcript version of this transcript. Bigger version numbers are later versions.	N/A	N/A

Table Constraints:

Constraint Name	Constraint Description	Columns Related
notexists	Each interaction can have multiple transcriptions of the same type, but they must be marked as different versions.	InteractionID, cTranscriptType, Version