Markup Language
Last updated
Last updated
In , the input can include markups to explicitly define phonetic mappings for one or more words, allowing you to bypass the Speechace lexicon. This feature is supported for both en-US and en-GB requests and can be utilized in scoring, multiple-choice, or Validate Text requests.
Each markup has the pattern: [l1|l2|...|ln]{s1|s2|...|sn}, where:
l1, l2, ..., ln are substrings of a word in the input text, and
s1, s2, ..., sn are syllables of the word corresponding to l1, l2, ..., ln.
Each syllable, si, has this pattern: p1 p2 ... pk, where pi is a phoneme in for the current dialect (i.e. en-us or en-gb).
If pi is a vowel phoneme, pi ends with 0, 1 or 2 to denote the stress level of the syllable.
There should be at most one vowel per syllable.
You can now pass the parameter markup_language=arpa_mark to the and include markup for words within the text as follows:
In the text above, the words "read" and "fragments" are marked up to disambiguate them from heteronyms.
Optionally, markup can specify letter-to-phoneme mapping. This allows the Speechace API to return in the results for the marked-up words.
To add letter information in markup:
Prefix the first syllable: Add 2| before s1 (the first syllable) as follows: [l1|l2|...|ln]{2|s1|s2|...|sn}. This indicates that the markup contains phoneme-to-letter mapping information.
Mapping Syntax: Use zero or more pairs of angular brackets < and > to group substrings in li (letters) and si (phonemes) for establishing one-to-one correspondence when necessary.
li consists of a sequence of characters c1c2...cp
si consists of a sequence of phonemes p1 p2 ... pk
Simple Mapping: If there is a one-to-one mapping between letters and phonemes, no additional markup is needed.
Inference of Mapping: If there are no pairs of < and > in li and si, and if p (the number of phonemes) equals k (the number of letters), we infer:
c1 is mapped to p1
c2 is mapped to p2
...,
cp is mapped to pk
Complex Cases: For cases where character-to-phoneme mapping is more complex than one-to-one:
Use < and > to group adjacent characters and/or phonemes.
Empty brackets <> can be used to denote mappings where m characters correspond to n phonemes (with m >= 0 and n >= 0, but not both zero).
This transforms li and si into sequences of the same length to facilitate one-to-one mapping.
This can be quite complex, so let's provide illustrative examples that cover all real-world cases. This way, you can easily copy and modify an example instead of constructing the notation from scratch.
Here’s a polished and professional version of your examples:
Markup: [car]{2|k aa1 r}
Explanation:
This example features one syllable without the need for < and >. Here, l1 is "car" and s1 is "k aa1 r." We can infer that:
c1 maps to p1
c2 maps to p2
c3 maps to p3
Thus, we have:
c1 = c
c2 = a
c3 = r
p1 = k
p2 = aa1
p3 = r
Markup: [bi|li<ng>|<>ual]{2|b ay0|l ih1 ng|g w ah0 l}
Explanation:
In this case, we have zero characters mapping to one phoneme in l3 as <>ual and s3 as g w ah0 l. The <> allows us to infer that the phoneme g is mapped to no character. Thus:
c1 =
c2 = u
c3 = a
c4 = l
p1 = g
p2 = w
p3 = ah0
p4 = l
Markup: [M<>|<><r.>]{2|m <ih1 s>|t er0}
Explanation:
This example shows zero characters mapping to two phonemes. Here, l1 is M<> and s1 is m <ih1 s>. The <> and <ih1 s> allow us to infer that the phonemes ih1 s are mapped to no character. Thus:
c1 = M
p1 = m
p2 = ih1 s
Markup: [ear]{2|iy1 <> r}
Explanation:
In this instance, we have one character mapping to zero phonemes. Here, l1 is "ear" and s1 is "iy1 <> r." The <> indicates that the character a is mapped to no phoneme. Thus:
c1 = e
c2 = a
c3 = r
p1 = iy1
p2 =
p3 = r
Markup: [box]{2|b aa1 <k s>}
Explanation:
This example features one character mapping to two phonemes. Here, l1 is "box" and s1 is "b aa1 ." The <k s> allows us to infer that the character x is mapped to the phonemes k s. Thus:
c1 = b
c2 = o
c3 = x
p1 = b
p2 = aa1
p3 = k s
Markup: [si<gh>]{2|s ay1 <>}
Explanation:
In this case, we have two characters mapping to zero phonemes. Here, l1 is si<gh> and s1 is s ay1 <>. The <gh> and <> allow us to infer that the characters gh are mapped to no phoneme. Thus:
c1 = s
c2 = i
c3 = gh
p1 = s
p2 = ay1
p3 =
Markup: [no<th>|i<ng>]{2|n ah1 th|ih0 ng}
Explanation:
This example features two characters mapping to one phoneme. Here, l1 is no<th> and s1 is n ah1 th. The <th> allows us to infer that the characters th are mapped to the phoneme th. Thus:
c1 = n
c2 = o
c3 = th
p1 = n
p2 = ah1
p3 = th
Markup: [qu<eue>]{2|k <y uw1> <>}
Explanation:
In this instance, we have three characters mapped to zero phonemes. Here, l1 is qu<eue> and l2 is k <y uw1> <>. The <eue> and <> allow us to infer that the characters eue are mapped to no phoneme. Thus:
c1 = q
c2 = u
c3 = eue
p1 = k
p2 = y uw1
p3 =