Mandarin courses with speech recognition

[marksealey]

Complete beginner begs forgiveness for what must be a common question: is there a good course of Mandarin that makes use of speech recognition?

That is, one where I can use my USB headset and microphone to practice pronunciation.

So computer-based - or online; and Mac at that?

TIA!

[kudra]

try the thread on Best Study tools

[marksealey]

Thanks, kudra; I did actually scour that thread when I first registered here.

Nothing jumped out at me, though.

What I want is probably software (for Mac), or possibly a website, where I hear the original, speak/reproduce it in my microphone and then get corrections based on how my response has been analysed.

I'm told R/Stone has this - but have not read too many +ve reviews of that course.

Perhaps I'm asking too much?

Thanks again

[kudra]

I hear the original, speak/reproduce it in my microphone and then get corrections based on how my response has been analysed.

...

Perhaps I'm asking too much?

probably. Although what you describe might be a good product. You could have stored the 5 most common pronunciation errors for a given syllable, with software able to distinguish between acceptable and one of the errors, and then provide canned suggestions.

For careful work on pronunciation especially tones try the introductory pronunciation tape at fsi-language-courses.com for Chinese (mandarin).

Still it will be up to you to train yourself to hear what sounds you yourself are producing and hear errors. IMO.

[lokki]

Rosetta Stone and another one of the drilling softwares (might have been the "Tell me More" one) have it. They are next to useless though.

A lot of people, myself included, have found shadowing techniques the most effective way to develop pronunciation and accent. Apart from that, recording yourself and comparing to the native material is also a good way.

[marksealey]

kudra,

You could have stored the 5 most common pronunciation errors for a given syllable, with software able to distinguish between acceptable and one of the errors, and then provide canned suggestions.

That's exactly what I am thinking of, Yes

Thanks for the fsi-language-courses link. Very helpful!

When you say:

it will be up to you to train yourself to hear what sounds you yourself are producing

is that because you don't think speech recognition is accurate enough (yet)? Or because there simply isn't any software available to do that?

I saw a Rosetta Stone ad. on TV the other day which implied that they do use some sort of vocal feedback (a microphone for sure) - maybe it's just Yes or No in response to questions?

Again - your help much appreciated…

[marksealey]

Hi lokki,

I feel I should know what 'shadowing techniques' are - but don't.

I like the idea of recording myself and listening to it (me!) back

Is there a quick answer as to why RS and TmM are not recommended?

I'm starting with NPCR.

Thanks so much!

[roddy]

Download Audacity (search on here and Google to find it) and you can start copying and pasting bits of audio around in the same way you do with text in Word. Once you can do that you can compare very closely your pronunciation and a model from your textbook audio, whatever.

Don't have anything to add on speech recognition, bar this, which might be useful for tones, and to say that even speech recognition software that attempts to understand what you're saying in your native language is flawed - try reading a speech-software dictated letter from someone who hasn't trained the software on their own voice. Given that, I'm dubious about the ability of current technology to analyze your voice and give you useful feedback.

You could start trawling through links like these, but I suspect a good introduction to phonetics, some samples of native speaker pronunciation, Audacity, if at all possible feedback from native speakers and experienced learners (you can upload small sound files here) and lots of patience are all you need.

[marksealey]

roddy,

Thanks so much. Your help in pointing me in some interesting directions very much appreciated

Shall first look at Audacity (here).

I use MacSpeech Dictate (here).

As a complete beginner, though with a fair experience of European languages and very determined once I start, I reckon I can do well. Just want to make sure that I begin with - as most of you are implying - the best tones I can.

Thanks again!

[kudra]

Just want to make sure that I begin with - as most of you are implying - the best tones I can.

Don't get me started. I think you have the right idea. Anyway, see this old thread.

is that because you don't think speech recognition is accurate enough (yet)? Or because there simply isn't any software available to do that?

I think it is one thing to look at a sonogram of your pronunciation to tell if you are getting the tone right -- and another thing to be able to hear if you are making a mistake as you produce the syllable. The thing is I remember people who could consistently pick out which tone is spoken to them, but couldn't produce the correct tone. They couldn't listen critically to themselves while they were producing the sound. I don't think audacity will help with this real-time problem as the time delay between production and checking the sonogram might be too long.

I do agree that using audacity to cut and paste your audio next to correctly pronounced audio can be useful, although I never did it that way. Ideally you will have a machine passing judgment on your tone production, which is good.

I have used audacity, but no other language software.

The yale david and helen site has pronunciation drills that compare and practice tone combinations. I'll let you search for the link here on chinese-forums.

[marksealey]

Thanks again, Kudra!

Shall read through the Better Tones thread carefully

At first search I found a book for intermediate learners for "The yale david and helen site". There's more?

[kudra]

true, david and helen is a 2nd year college book, but this site has some pronunciation practice specific to tones and tone combinations. It looks like all of the Fall tutorial links have similar tone practice.

[marksealey]

kudra,

Yes, thanks. Shall start looking at its root, here.

[lokki]

is that because you don't think speech recognition is accurate enough (yet)? Or because there simply isn't any software available to do that?

Basically I think it's the recognition software that is not accurate, or clever enough yet, at least not the ones incorporated in these programs.

What you get is a graphic curve of the pitch pattern of the native recording and then a similar curve of your own voice input and you can compare the two curves. In addition you get an evaluation of how close you got, represented as a circular guage (similar to a classic car speedometer). When the software thinks you got close enough the guage will end up in the "green" area at the right-hand end of the scale.

All this would be fine, except that in my experience - and I have read reviews of others reaching the same conclusions - it is very questionable whether the software really measures the quality of your pronunciation, or if it measures something else. I found I could almost never get an "approved" result (in the green arc) using my normal voice, but if I affected a squeaky high-pitched shriek I got green nearly every time. Experimenting further, I could also get an approved result while saying some completely different words, with the same intonation pattern. If I had continued using, and trusting that software I think I would just have developed some very artificial and unnatural pronunciation habits - to satisfy the machine, but I wouldn't necessarily have been any better at making myself understood to native speakers.

There might be some value in playing around with the interface initially and comparing the curves, but I'd take those results with a big pinch of salt and move on to shadowing and recording as soon as possible, for more serious accent training work.

I didn't explain shadowing since it has been discussed before, here and elsewhere. A search for "shadowing" should turn up plenty of information.

Edited September 22, 2008 at 05:05 PM by lokki

[marksealey]

Thanks, Lokki - all very clear.

My next step is to try Audacity out in depth and in place.

Is there a difference between 'shadowing' and merely repeating afterwards?

Your help and guidance appreciated

[roddy]

I was actually just about to post and say that I'd never seen the difference between shadowing, and copying. . .

[marksealey]

Roddy, Yes. I used to teach E'F'L (as it was then), then ESL, in the '70's. We always used to say 'repeat'…

[lokki]

The terminology is not very consistent on these things but as I understand it, repeating is speaking after the recording: Recording says: "Good morning", followed by a three-second silence for you to repeat "Good Morning".

Shadowing means speaking at the same time as the recording - or as close to it as you can manage, and repeating lots of times. You may be half a second behind initially, but after repeating the same short phrase a few dozen times you can do it simultaneously with the native voice and gradually your pitch and other patterns melt down and adapt to the native patterns as to a mold.

Another term used is chorusing, which would be a group of people, such as a whole class full of students, reciting phrases in unison. To some people shadowing and chorusing is the same thing.

EDIT: I don't think I have come across "copying" in this context and I am not sure where it would fit in. All the different techniques have an element of copying and imitation, but they are still quite different approaches.

Edited September 22, 2008 at 08:31 PM by lokki

[marksealey]

Thanks, lokki: all makes good sense.

[jbradfor]

What I assume the OP is interested in, as I am, is something like the following, except for Mandarin and not English

Even as Indian call centers have thrived in the past decade, helping U.S. companies cut costs and creating hundreds of thousands of jobs in India, they have faced a seemingly insurmountable problem: Most Indian employees speak heavily accented English.

Now IBM Corp.'s India Research Lab says it has a way to help operators fix the harsh consonants, local idioms and occasionally different grammar of Indian English, often a source of frustration of those who call in search of tech support and other information.

IBM, which operates large call center facilities in India, has developed a Web-based training technology that can help improve the language skills of operators.

....

The program evaluates grammar, pronunciation, comprehension and other spoken-language skills, and provides detailed scores for each category. It uses specially adapted speech-recognition software to score the pronunciation of passages and the stressing of syllables for individual words.

The technology also consists of voice-enabled grammar evaluation tests, which identify areas for improvement by highlighting shortcomings and providing examples of correct pronunciation and grammar.

http://www.cbsnews.com/stories/2006/11/07/tech/main2160939.shtml

I emailed them to see if they were coming out with other languages (e.g. Mandarin) anytime. They said no, that this software relies on speech recognition as its basis, which needs to be done separately for each language.

But I'm still hoping some other company will come out with a Mandarin version.