Education and demonstration – Language – Spelling – phonics – word recognition – or sentence formation
Patent
1991-02-25
1996-04-02
Apley, Richard J.
Education and demonstration
Language
Spelling, phonics, word recognition, or sentence formation
434156, 434157, 434169, 434178, 434185, G09B 1900
Patent
active
055035604
DESCRIPTION:
BRIEF SUMMARY
This invention relates to apparatus and methods for training pronunciation; particularly, but not exclusively, for training the pronunciation of second or foreign languages.
One type of system used to automatically translate speech between different foreign languages is described in our European published patent application number 0262938A. This equipment employs speech recognition to recognise words in the speaker's utterance, pattern matching techniques to extract meaning from the utterance and speech coding to produce speech in the foreign tongue.
This invention uses similar technology, but is configured in a different way and for a new purpose, that of training a user to speak a foreign language.
This invention uses speech recognition not only to recognise the words being spoken but also to test the consistency of the pronunciation. It is a disposition of novice students of language that, although they are able to imitate a pronunciation, they are liable to forget, and will remain uncorrected until they are checked by an expert. A machine which was able to detect mispronunciation as well as translation inaccuracies would enable students to reach a relatively high degree of proficiency before requiring the assistance of a conventional language teacher to progress further. Indeed, very high levels of linguistic skill are probably not required in the vast majority of communication tasks, such as making short trips abroad or using the telephone, and computer aided language training by itself may be sufficient in these cases.
Conventional methods either involve expensive skilled human teachers, or the use of passive recordings of foreign speech which do not test the quality of the student's pronunciation.
Some automated systems provide a visual display of a representation of the student's speech, and the student is expected to modify his pronunciation until this display matches a standard. This technique suffers from the disadvantage that users must spend a great deal of time experimenting and understanding how their speech relates to the visual representation.
Another approach (described for example in Revue de Physique Appliquee vol 18 no. 9 Sept 1983 pp 595-610, M. T. Janot-Giorgetti et al, "Utilisation d'un systeme de reconnaissance de la parole cornroe aide a l'acquisition orale d'une langue etrangere") employs speaker independent recognition to match spoken utterances against standard templates. A score is reported to the student indicating how well his pronunciation matches the ideal. However, until speaker independent recognition technology is perfected, certain features of the speaker's voice, such as pitch, can affect the matching scores, and yet have no relevant connection with the quality of pronunciation. A student may therefore be encouraged to raise the pitch of his voice to improve his score, and yet fail to correct an important mispronunciation.
Furthermore, current speaker independent recognition technology is unable to handle more than a small vocabulary of words without producing a very high error rate. This means that training systems based on this technology are unable to process and interpret longer phrases and sentences. A method of training pronunciation for deaf speakers is described in Procedings ICASSP 87 vol 1 pp 372-375 D. Kewley-Port et al `Speaker-dependant Recognition as the Basis for a Speech Training Aid`. In this method, a clinician selects the best pronounced utterances of a speaker and these are convened into templates. The accuracy of the speaker's subsequent pronunciation is indicated as a function of his closeness to the templates (the closer the better). This system has two disadvantages; firstly, it relies upon human intervention by the clinician, and secondly the speaker cannot improve his pronunciation over his previous best utterances but only attempt to equal it.
According to the invention there is provided apparatus for pronunciation training comprising;
speech generation means for generating utterances; and
speech recognition means arranged to recognise in a
REFERENCES:
patent: 3703602 (1972-11-01), Shensheu
patent: 3946504 (1976-03-01), Nakano
patent: 4507750 (1985-03-01), Frantz et al.
patent: 4615680 (1986-10-01), Tomatis
patent: 4884972 (1989-12-01), Gasper
Apley Richard J.
British Telecommunications
Richman Glenn E.
LandOfFree
Language training does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Language training, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Language training will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2013335