nzilbb.formatter.text
nzilbb.formatter.text (1.5.1)
Serializer/Deserializer for transcripts in plain text format.
Support for plain-text transcripts includes both import and export of transcripts as .txt files. Transcripts can be textual documents, or transcriptions of speech attributable to different speakers.
For importing/parsing files, the document structure is assumed to be a meta-data ‘header’ followed by speaker turns introduced by speaker IDs. Each line in the speaker turn is taken to be an utterances.
The speaker ID of a turn is a word followed by a colon, by default (this is configurable); that marks the beginning of utterances by that speaker, and all subsequent lines are assumed to belong to that speaker/turn, until the next speaker ID is encountered.
All lines before the first speaker ID are assumed to be part of the ‘header’, each line possibly containing meta-data of the form key=value (this form is also configurable). Header lines that don't conform to the meta-data pattern are prepended to the transcript as comment annotations.
If there are no speaker IDs found, the entire transcript is assumed to be a text rather than a transcript, and all lines are attributed to a single person called “author”.
Synchronization
Utterances can be synchronized to a separate recording by including time-stamp lines throughout the text.
A time-stamp must occur on a line by itself, and is assumed to use the
following format by default (the format is configurable):
HH:mm:ss.SSS
If time-stamps are found in the tex, or it is loaded with a sound file, the document is assumed to be a transcript of a recording, and offsets are in seconds. Otherwise, the document is assumed to be a written text, and offsets are in characters.
Annotation Conventions
A number of textual annotation conventions are supported:
- Non-speech commentary
- enclosed in curly braces (surrounded by spaces), e.g.
the he went {pointing} over there - Noises
- enclosed in square brackets (surrounded by spaces), e.g.
now [door slamming] let's start - Pronunciation
- enclosed in square brackets, immediately following the word it's
annotating, with no intervening space, e.g.
this was at Wingatui[wIN@tui] - Lexical tags
- enclosed in round parentheses, immediately following the word it's
annotating, with no intervening space, e.g.
I can't remem~(remember)
Configuration
The following parameters can be specified for the formatter:
- participantLayer
- Layer for speaker/participant identification (“participant” by default)
- turnLayer
- Layer for speaker turns (“turn” by default)
- utteranceLayer
- Layer for speaker utterances or text lines (“utterance” by default)
- wordLayer
- Layer for individual word tokens (“word” by default)
- commentLayer
- Annotation layer for commentary (“comment” by default)
- noiseLayer
- Annotation layer for Background noises (“noise” by default)
- lexicalLayer
- Word annotation layer for lexical tags (“lexical” by default)
- pronounceLayer
- Word annotation layer for non-standard pronunciation tags (“pronounce” by default)
- orthographyLayer
- Word token orthography layer (excluding punctuation etc.) used for when exporting texts with other word tags (“orthography” by default)
- useConventions
- Whether to use text conventions for comment, noise, lexical, and pronounce annotations
- maxParticipantLength
- The maximum length of a participant name/ID, when starting a speaker turn (20 characters by default)
- maxHeaderLines
- The maximum number of lines in a meta-data header (50 by default)
- participantFormat
- Format for marking a change of turn within the transcript body -
e.g.
{0}:, where{0}is a place-holder for the participant ID/name - metaDataFormat
- Format for a meta-data line in the header - e.g.
{0}={1}, where{0}is a place-holder for the attribute name or key, and{1}is a place-holder for the attribute value - timestampFormat
- Format for a time stamp - e.g.
HH:mm:ss.SSS
Exporting texts with word tags
If an annotation graph is exported with an orthography layer and one
or more word-tag layers (e.g. POS tags, etc.), then in the output
text, each word token has its annotation tags appended, with
underscore _ as the delimiter between token/tags.