Skip to Content

Department of Linguistics

TTS: The Disambiguation of Homographs and Stress Alternation Word Pairs

Robert Mannell

In English there are numerous word pairs that have the same spelling, but different pronunciation. These words are a problem for TTS systems. A TTS system needs to make a decision as to which pronunciation should be applied to such words. On the other hand, homophones (words with different origin or meaning, the same or different spelling, but the same pronunciation) are not a problem as there is no need to select between different potential pronunciations.

There are several types of homograph in English:-

  1. Words of differing origin and meaning, but the same spelling. This group can be divided into two sub-groups (from the perspective of a TTS system).
    1. Words with the same spelling and the same pronunciation (eg. bank vs. bank, fair vs. fair). These words are not a problem for TTS systems.
    2. Words with the same spelling, but different pronunciation (eg. bow /b@u/ vs. bow /bau/). Such words are a very big problem for TTS systems.
  2. Words related semantically but whose pronunciation varies, typically with respect to stress placement, depending on the gramatical role (part-of-speech) that the word is playing in the sentence (eg. 'record [noun] vs. re'cord [verb]). These words are also a big problem for TTS systems.

The two main types of homophones that cause difficulties for TTS systems require quite different strategies.

Homographs of type 1a require some mechanism for determining the intended meaning of the word, whereas homographs of type 2 require a method for determining the part-of-speech of the word. In other words, one class of homophones requires a semantic solution and the other class requires a syntactic solution.

A List of Some Homographs and Words with Grammatical Stress Alternations

This list represents less than 7% of the total number of words of this kind in the current MU-Talk TTS dictionary

absent| %hom. #1 \P ['Abs@nt] \G adj. \R (absent) #2 \P [@b'sEnt] \G adj. \R (absent)
abstract| %hom. #1 \P ['AbstrAkt] \G adj. n. \R (abstract) #2 \P [@b'strAkt 'AbstrAkt] \G v. \R (abstract)
abuse| %hom. #1 \P [@'bju:z] \G v. \R (abuse) #2 \P [@'bju:s] \G n. \R (abuse)
address| %hom. #1 \P [@'drEs 'AdrEs] \G n. \R (address) #2 \P [@'drEs] \G v. \R (address)
advocate| %hom. #1 \P ['Adv@keit] \G v. \R (advocate) #2 \P ['Adv@k@t 'Adv@keit] \G n. \R (advocate)
affix| %hom. #1 \P [@'fIks] \G v. \R (affix) #2 \P ['AfIks] \G n. \R (affix)
appropriate| %hom. #1 \P [@'pr@upri:@t] \G adj. \R (appropriate) #2 \P [@'pr@upri:eit] \G v. \R (appropriate)
approximate| %hom. #1 \P [@'prOks@m@t] \G adj. \R (approximate) #2 \P [@'prOks@meit] \G v. \R (approximate)
associate| %hom. #1 \P [@'s@uSi:eit @'s@usi:eit] \G v. \R (associate) #2 \P [@'s@uSi:@t @'s@usi:@t] \G n. adj. \R (associate)
august| %hom. #1 \P [o:'gVst] \G adj. \R (august) #2 \P ['o:g@st] \G n. \R (august)
bow| %hom. #1 \P [bau] \G v. n. \R (bow) #2 \P [b@u] \G n. adj. v. \R (bow)
combine| %hom. #1 \P [k@m'bain] \G v. \R (combine) #2 \P ['kOmbain] \G n. \R (combine)
compound| %hom. #1 \P ['kOmpaund] \G adj. n. \R (compound) #2 \P [k@m'paund] \G v. \R (compound)
compress| %hom. #1 \P [k@m'prEs] \G v. \R (compress) #2 \P ['kOmprEs] \G n. \R (compress)
concert| %hom. #1 \P ['kOns@t] \G n. \R (concert) #2 \P ['kOns@:t] \G n. \R (concert) #3 \P [k@n's@:t] \G v. \R (concert)
conduct| %hom. #1 \P ['kOndVkt] \G n. \R (conduct) #2 \P [k@n'dVkt] \G v. \R (conduct)
content| %hom. #1 \P ['kOntEnt] \G n. \R (content) #2 \P [k@n'tEnt] \G adj. v. n. \R (content)
convert| %hom. #1 \P [k@n'v@:t] \G v. \R (convert) #2 \P ['kOnv@:t] \G n. \R (convert)
convict| %hom. #1 \P [k@n'vIkt] \G v. \R (convict) #2 \P ['kOnvIkt] \G n. \R (convict)
decrease| %hom. #1 \P [d@'kri:s] \G v. \R (decrease) #2 \P ['di:kri:s d@'kri:s] \G n. \R (decrease)
delegate| %hom. #1 \P ['dEl@g@t 'dEl@geit] \G n. \R (delegate) #2 \P ['dEl@geit] \G v. \R (delegate)
deliberate| %hom. #1 \P [d@'lIb@r@t] \G adj. \R (deliberate) #2 \P [d@'lIb@reit] \G v. \R (deliberate)
desert| %hom. #1 \P ['dEz@t] \G n. adj. \R (desert) #2 \P [d@'z@:t] \G v. n. \R (desert)
elaborate| %hom. #1 \P [@'lAb@r@t] \G adj. \R (elaborate) #2 \P [@'lAb@reit i:'lAb@reit] \G v. \R (elaborate)
estimate| %hom. #1 \P ['Est@meit] \G v. \R (estimate) #2 \P ['Est@m@t] \G n. \R (estimate)
frequent| %hom. #1 \P ['fri:kw@nt] \G adj. \R (frequent) #2 \P [fr@'kwEnt] \G v. \R (frequent)
graduate| %hom. #1 \P ['grAdZu:@t] \G n. adj. \R (graduate) #2 \P ['grAdZu:eit] \G v. \R (graduate)
impact| %hom. #1 \P ['ImpAkt] \G n. \R (impact) #2 \P [Im'pAkt] \G v. \R (impact)
import| %hom. #1 \P [Im'po:t 'Impo:t] \G v. \R (import) #2 \P ['Impo:t] \G n. \R (import)
lead| %hom. #1 \P [li:d] \G v. n. adj. \R (lead) #2 \P [lEd] \G n. v. adj. \R (lead) ! (lead <1> %suf -ed)
legitimate| %hom. #1 \P [l@'dZIt@m@t] \G adj. n. \R (legitimate) #2 \P [l@'dZIt@meit] \G v. \R (legitimate)
live| %hom. #1 \P [lIv] \G v. \R (live) #2 \P [laiv] \G adj. adv. \R (live)
minute| %hom. #1 \P ['mIn@t] \G v. n. adj. \R (minute) #2 \P [mai'nju:t] \G adj. \R (minute)
overall| %hom. #1 \P ['@uv@ro:l] \G adj. n. \R (all %pre over-) #2 \P [@uv@r'o:l] \G adv. \R (all %pre over-)
perfect| %hom. #1 \P ['p@:f@kt] \G adj. n. \R (perfect) #2 \P [p@:'fEkt p@'fEkt] \G v. \R (perfect)
progress| %hom. #1 \P ['pr@ugrEs] \G n. \R (progress) #2 \P [pr@'grEs] \G v. \R (progress)
project| %hom. #1 \P ['pr@udZEkt 'prOdZEkt] \G n. \R (project) #2 \P [pr@'dZEkt] \G v. \R (project)
record| %hom. #1 \P [r@'ko:d] \G v. \R (record) #2 \P ['rEko:d] \G n. adj. \R (record)
refund| %hom. #1 \P ['ri:fVnd] \G n. \R (fund %pre re-) #2 \P [r@'fVnd ri:'fVnd] \G v. \R (fund %pre re-)
refuse| %hom. #1 \P [r@'fju:z] \G v. \R (refuse) #2 \P ['rEfju:s] \G n. adj. \R (refuse)
reject| %hom. #1 \P [r@'dZEkt] \G v. \R (reject) #2 \P ['ri:dZEkt] \G n. \R (reject)
separate| %hom. #1 \P ['sEp@reit] \G v. \R (separate) #2 \P ['sEpr@t] \G adj. n. \R (separate)
suspect| %hom. #1 \P [s@'spEkt] \G v. \R (suspect) #2 \P ['sVspEkt] \G n. adj. \R (suspect)
transfer| %hom. #1 \P [trAns'f@:] \G v. \R (transfer) #2 \P ['trAnsf@:] \G n. \R (transfer)
transform| %hom. #1 \P [trAns'fo:m] \G v. \R (form %pre trans-) #2 \P ['trAnsfo:m] \G n. \R (form %pre trans-)
transport| %hom. #1 \P [trAns'po:t 'trAnspo:t] \G v. \R (transport) #2 \P ['trAnspo:t] \G n. adj. \R (transport)
use| %hom. #1 \P [ju:z] \G v. \R (use) #2 \P [ju:s] \G v. n. \R (use)
wind| %hom. #1 \P [wInd] \G n. v. \R (wind) #2 \P [waind] \G v. n. \R (wind)
wound| %hom. #1 \P [wu:nd] \G n. v. \R (wound) #2 \P [waund] \G v. adj. \R (wind <2> %suf -ed)

Bibliography

  1. Yarowski D., (1997) "Homograph disambiguation in Text-to-Speech Synthesis", In van Santen J.P.H., Sproat R.W., Olive J.P., and Hirschberg J., (eds.) Progress in Speech Synthesis, Springer, New York, pp157-172