voice recognition program for Thais learning English

[troysantos]

I know there are programs out there that will show you sentences to read and then record your voice. Then compare your pronunciation with the same text spoken by a native speaker. Of course, no two native speakers (even from the same hometown or even siblings!) say everything exactly alike. So, comparing a language learner's pronunciation with the program is open to criticism, but I think it's useful just the same.

So, anyone know of such a program? I only know that there are such programs but I don't know any names.

)

[zeichen]

There is a book series called touchstone that has it attached to their students book. The problem is that it is preset with the dialogs. I would love to have the raw program that allowed teachers to record their voices and then have students practice at home as a self study thing. I was making some interactive story books and wanted that to be a feature also but after spending a lot of time and some money getting a software developer to create it, I gave up.

[zeichen]

There is a book series called touchstone that has it attached to their students book. The problem is that it is preset with the dialogs. I would love to have the raw program that allowed teachers to record their voices and then have students practice at home as a self study thing. I was making some interactive story books and wanted that to be a feature also but after spending a lot of time and some money getting a software developer to create it, I gave up.

I know that buying the full program from the few major retailers costs several thousand dollars.

[troysantos]

)

[Scott]

A troll post has been removed.

[Eliot Rosewater]

That is a great idea! I often use S Voice on my samsung to write emails; it works pretty well but can sometimes mangle even a native speaker's message (I think there is a whole website dedicated to funny mistakes on this)

The iPhone equivalent is Siri; many of your students probably have one or the other.

A cool activity would be for you to hand out a written message which they must then dictate by voice and get the message right on their device.

[lostinisaan]

Again, the program "Tell me More" seems to be worth a try. P.S. This is not a troll.

[brianp0803]

Have you tried rosetta stone?

They have different languages and you can say words and it has mechanical representative of your voice and native speaker's voice.

[harrry]

An even better way is to use a windows 8.1 mobile phone, It has a translate option. The person speaks in English and it writes it out in very acceptible Thai. If his English is good enough it will understand him. If not it will not.

[zeichen]

sorry translator programs don't help people build their pronunciation or listening skills. I think that you missed the point.

[harrry]

sorry translator programs don't help people build their pronunciation or listening skills. I think that you missed the point.

No I got the point exactly.

If they can speak into the phone and it correctly recognises what they say they have mastered the pronunciation to a usable level. If they cannot it is unlikely that they are speaking correctly. I am not talking about a program to translate to English. I am referencing a program which accepts a well spoken English voice input and outputs the text of what you said in Thai.

Edited September 27, 2014 by harrry

[Zooheekock]

I've used Dragon Dictate for myself (just out of curiosity) which I think is about as good as any voice recognition app. The problem with it, as a teaching tool, is that you have to pronounce words more-or-less in isolation, suppressing normal processes like elision and liaison and this probably encourages some bad habits; Thai learners tend to over-produce their words and voice recognition software may make that even worse. If you want to do more detailed comparisons, you can use Praat (free to download from http://www.fon.hum.uva.nl/praat/) but you need to know a little about phonetics to use it.

[JDGRUEN]

I have had the idea of using the software 'Dragon Naturally Speaking' (DNS) as a language teaching adjunct. DNS is an excellent voice to text conversion software ... not cheap - but can be downloaded... While not meant for language teaching nor accent reduction, I believe it could be used for these purposes....

Using DNS for its intended purpose - in normal mode for dictation - it is a short cut to bypass typing -- just speak into the microphone and text appears on the screen... It is quite good and can be downloaded also as a Smartphone app... DNS is set or can be set to become accustom to the speakers tone and accent - it learns to become better at understanding the individual speaker's words and more accurately converting them to written English. I have seen it in action and it is quite accurate... There are many youtube video demonstrating Dragon Naturally Speaking. One has a man with a noticeable Texas accent working with DNS to dictate and it discusses how the software adjusted to his accent. I think the reverse could be set - not to adjust but to expect non accented English.

I believe the learning mode of DNS could be turned off to where it is expecting to hear a generic American English speaker - and not adapt to your tone, pace, prounciation and accent. If this is the case - then DNS could be set to expect to 'hear' that generic American English speaker ... and would convert the spoken word incorrectly until the word(s) are 'heard' intelligibly.

So - theoretically one could sit in complete privacy with a head set and a laptop or tablet or whatever... Load up DNS -- and start dictating an English language practice sentence and one could keep repeating the pronunciation and enunciation - elocution ? until the word(s) are spelled correctly on the screen... This would give instant feed back from a non human - non critical source... no joking around ... just a type of written echo of what is 'heard' as translated - converted from voice to text. Just keep the DNS 'learning' mode turned off so DNS will not to become accustom to your voice and accent and allowed to keep on expecting to hear a midwest accent generic American accent...

I am not 100% positive that DNS could be set to work this way -- but it is worth a shot to find out. If it can - I think it has great potential to teach shy Thai people to learn the English - as an adjunct - not as a replacement for all sorts of other methods. I have been told by many Thais that classroom criticism of their speaking efforts is more than they can deal with - so they do not get into spoken English due to embarrassment and shyness.

[Zooheekock]

As I said above, the software may be very impressive in its primary role but if you try to teach pronunciation with something like Dragon, (i) you're encouraging learners to produce non-standard speech and (ii) it does nothing to help with areas where Thai learners have real problems, which is really with things like understanding and producing syllabic consonants, vowel weakening, syllable-linking, word stress, phrasing, tonic stress, etc (and because of (i) it may make those worse). I'm also not sure that just at the level of individual sounds, it's going to be that much good. When Dragon differentiates between 'shoe' and 'chew', for example, is it doing that purely on acoustic features of the input or is it doing it on relative distributions of the words within sentences? So if I say 'That's a lovely pair of shoes' it's pretty much impossible to misunderstand that for 'That's a lovely pair of chews' just because the second sentence is never going to appear in any natural input. I suspect (though I confess, I am guessing here), some of what the software does is look at these types of features so students may, perhaps, be able to say 'That's a lovely pair of chews' but still be marked correct because the software is guessing the closest probable output.

That said, there are cases where it could be useful. For practicing production of individual words, it might have a role. I can see that sitting there, repeating 'Sue' and 'zoo' until you can reliably hit the right one could be productive, though even there, you would have to be a little careful - I can also see that students sitting there trying to get 'mob' and 'mop' right might end up massively over-producing the release of the final consonant.

Edited September 27, 2014 by Zooheekock

[harrry]

As I said above, the software may be very impressive in its primary role but if you try to teach pronunciation with something like Dragon, (i) you're encouraging learners to produce non-standard speech and (ii) it does nothing to help with areas where Thai learners have real problems, which is really with things like understanding and producing syllabic consonants, vowel weakening, syllable-linking, word stress, phrasing, tonic stress, etc (and because of (i) it may make those worse). I'm also not sure that just at the level of individual sounds, it's going to be that much good. When Dragon differentiates between 'shoe' and 'chew', for example, is it doing that purely on acoustic features of the input or is it doing it on relative distributions of the words within sentences? So if I say 'That's a lovely pair of shoes' it's pretty much impossible to misunderstand that for 'That's a lovely pair of chews' just because the second sentence is never going to appear in any natural input. I suspect (though I confess, I am guessing here), some of what the software does is look at these types of features so students may, perhaps, be able to say 'That's a lovely pair of chews' but still be marked correct because the software is guessing the closest probable output.

That said, there are cases where it could be useful. For practicing production of individual words, it might have a role. I can see that sitting there, repeating 'Sue' and 'zoo' until you can reliably hit the right one could be productive, though even there, you would have to be a little careful - I can also see that students sitting there trying to get 'mob' and 'mop' right might end up massively over-producing the release of the final consonant.

It is much more useful for full sentences as it takes all normal rules into account and looks for the best match which makes sense, just as a normal listener does. It is really not that good for individual words as it cannot use context to determine if it has the right match or not and even in normal speech the pronunciation of each word is never perfect.

[JDGRUEN]

As I said above, the software may be very impressive in its primary role but if you try to teach pronunciation with something like Dragon, (i) you're encouraging learners to produce non-standard speech and (ii) it does nothing to help with areas where Thai learners have real problems, which is really with things like understanding and producing syllabic consonants, vowel weakening, syllable-linking, word stress, phrasing, tonic stress, etc (and because of (i) it may make those worse). I'm also not sure that just at the level of individual sounds, it's going to be that much good. When Dragon differentiates between 'shoe' and 'chew', for example, is it doing that purely on acoustic features of the input or is it doing it on relative distributions of the words within sentences? So if I say 'That's a lovely pair of shoes' it's pretty much impossible to misunderstand that for 'That's a lovely pair of chews' just because the second sentence is never going to appear in any natural input. I suspect (though I confess, I am guessing here), some of what the software does is look at these types of features so students may, perhaps, be able to say 'That's a lovely pair of chews' but still be marked correct because the software is guessing the closest probable output.

That said, there are cases where it could be useful. For practicing production of individual words, it might have a role. I can see that sitting there, repeating 'Sue' and 'zoo' until you can reliably hit the right one could be productive, though even there, you would have to be a little careful - I can also see that students sitting there trying to get 'mob' and 'mop' right might end up massively over-producing the release of the final consonant.

I think your guessing without actual experimentation is just - well ... guessing... I believe it would take considerable experimentation before any conclusion could be made.

However, your mental analysis does point to way such speech recognition software could be tailored in it's programming code to have language training scripts in a data file that streams across the screen and is the source of what the student is learning to speak... Students read streaming text while student is speaking the same text... the speech recognition software determines the proper pronunciation match by recognizing the spoken word. Play - attempt speaking - play again - attempt speaking.... This would be much the same as a language coach giving feedback - only difference the speech recognition software program would have infinitive patience for retries and mistakes and no possible harsh criticism. This programming approach would also allow single word pronunciation practice because the software program would already 'know' what word to expect. Perhaps the software would 'know' what to expect - not just from having the expected word in a data file - but how about - even having the students spoken word(s) come close in proximity to a waveform pattern of the expected word?

In other words this idealistic linguistic software I dream about would be a combination of speech recognition software enhanced and coupled with speech waveform analysis software and a waveform database. This is not so fantastical as it may sound. I have recently researched such different kinds of software and found several having some of these properties. One is advertised as a accent reduction aide.

I would suggest that an entrepreneurial team made up of language instructors, programming coding folks, people experienced in voice waveform analysis and related technologies could produce a great software product. First by getting a license to alter the code of a good speech recognition software (no need to reinvent the wheel. Then produce the end product with the combined effort of the team. I believe the market for such a product would be very large for such a unique product.

Pie in the sky" you say! ... Many successful products came about this very way... daydreaming a little with realistic concepts.

And pleae note -- when I proposed the original concept in my post, I used the word ADJUNCT... not a replacement for language training/instruction. This would be especially so for the many Thai people I have met who point out that their English language training in a Thai classroom was a grueling ordeal. Which I feel is why I have met a number of grown Thai women who can type, read and comprehend English quite well - but by their own admission cannot speak it very well at all. And I would have to say they are correct.

[JDGRUEN]

Borrowed ...

"Tea-CHAH!"

We've all heard this distinct Thai accent, but how can we as teachers help students realize that their pronunciation is so unusual, let alone to help correct it.

Students can use a new speech recognition app to hear a word or phrase in native English speech and then check their own pronunciation with a speech recognition API developed by MIT. It's free and runs on GOOGLE CHROME.

Here is a demo for a social studies unit on Galileo Galilei.

http://grade9ell.comeze.com

[Zooheekock]

^ I just tried that. You can completely mis-stress the words and they are still recognized as correct so I'm not sure that it's going to help much with problems such as /ˈti:ʧə/ vs /ti:ˈʧa:/ (besides, if you're a teacher, you should be doing this yourself, not relying on some IT substitution).

The problem (or at least one problem) with speech recognition is that it has a very different goal to accent training. If you're designing a speech recognition program, your goal is to make the recognition broad enough to encompass the differences in pronunciation which occur between speakers (and actually within the output of the same speaker), though not so broad as to mis-recognize words. Accent training has to pick one accent and force you as close as possible to that model. The former is naturally inclusive whereas the latter is naturally exclusive.