[postproc] Module Configuration

The postproc module filters and modifies results produced by audio processing tasks such as speech-to-text, language identification, and so on.

Input and Output

The postproc module has two modes of operation, which can also be combined to perform both processes in a single operation.

Mode Input Output Description
B w w Accepts a word data stream, and replaces all barred words either with a fixed term (such as <BLEEP>), or a term specific to an individual word. The barred word list is provided as a text file containing one word or a pair of words (the barred word, and the replacement term for that word) on each line.
R w w Accepts a data stream, and recombines any word fragments to form complete words. Writes the resulting word sequence to an output word stream.
P w1 w2

Include simple sentence-forming punctuation (for example, full stops and initial capital letters) in the output. HPE recommends that you do not use the output as the input for other modules; this option is designed purely for display purposes, and to produce more human-readable output.

This mode uses periods of silence in a sequence of words to break it up into sentences with added punctuation, but you must remove any periods of silence manually from the punctuated string.

NOTE:

You can use this mode in conjunction with word barring, but for use in conjunction with any other mode, you must call postproc once for each mode.

CAUTION:

This mode is designed for use with Latin languages, and therefore is not recommended for use with non-Latin languages.

_    

Carries out postprocessing of the output of a speech-to-text module, By default, no operation is performed for most languages.

This mode receives words from the speech-to-text process and feeds them without delay to the next process (usually the word output process). The module carries out word recompounding automatically if the language requires it (currently the only languages that require recompounding are TRTR and HUHU).

The following example shows a sample configuration file entry for this mode:

[postproc]
doPunctuation = $params.punctuation
doWordBarring = $params.wordBar
barredList = $params.wordBarList
lang = $stt.lang
forceRecompoundOn = $params.forceRecompoundOn

The Lang parameter is the language from the speech-to-text module. The decision on whether to recompound is based on this parameter (that is, whether the language is TRTR or HUHU). You can use the ForceRecompoundOn parameter to enforce recompounding for other languages. A comparable ForceRecompoundOff parameter is available to prevent recompounding on TRTR or HUHU, but in most operations this is not necessary.

Set the Punctuation parameter to add punctuation to the word data.

Set WordBar to True and specify a list of words to be barred as the value of the WordBarList parameter. For example, you could replace swear words with <BLEEP> or replace ATNT with AT&T.

You can perform the recombination, punctuation, and word barring modes simultaneously. You must request other postprocessing modes with a separate instance of postproc.

NOTE:

Cascading of postproc instances is permitted.

Examples:

w2 ← postproc (B,w1)
w2 ← postproc (R,w1)
w2 ← postproc (BR,w1)
W2 <- postproc(P, w1)
W2 <- postproc(_,...)

Parameters

NOTE:

In B mode, you must set at least one of the BarredList and BarredTerm parameters.

BarredList Punctuation
BarredTerm RcmpValidList
ForceRecompoundOff RcmpAllowSuffix
ForceRecompoundOn WordBar
NonSentFinalWords WordBarList

_HP_HTML5_bannerTitle.htm