Sanas makes emerging-market English sound American or British

Topic

New Reply

PUBLIC DEFENDER By Brian Livingston A tiny tech startup says its software can almost instantly convert the accent of an ESL (English as a Second Langu
[See the full post at: Sanas makes emerging-market English sound American or British]

Reply | Quote

Replies

Does this mean that the scammer based in (a big Indian city) who phones me to tell me that the Amazon Prime account that I don’t have has been charged £ 79.99 will now speak with impeccable RP (Received Pronunciation) English?  His almost unrecognisable accent in English was my initial defence against him.

Dell E5570 Latitude, Intel Core i5 6440@2.60 GHz, 8.00 GB - Win 10 Pro

4 users thanked author for this post.

Charlie, wavy, Cybertooth, OscarCP

Reply | Quote

We take the good with the bad, don’t we?

Counterpoint — I have a 3rd tier repackaged phone service for my cell phone. Their customer service is ESL. I’d kill for this type of ESL to Real American English conversion in those conversations! Similarly with my Cable Company’s Customer Service Reps.

So I guess I’ll have to find some other way to know that my FedEx package is or is not waiting for me to confirm my identity.

I think I’ll still know that Microsoft is not calling me to remove a virus they just detected on my Windows 10 PC — when I’m actually running Linux on my Chromebook at the time!

-- rc primak

2 users thanked author for this post.

Charlie, wavy

Reply | Quote

I think the scammers have caught on. I have observed that speakers in many scam calls nowadays are native speakers of English. i have had calls purportedly from the IRS, Amazon, Microsoft (you name it) in which the person sounded no different from David Muir or Lindsey Davis of ABC news.

Reply | Quote

Or perhaps the target market of this deceit are those who wish to conceal the origin of tech support idiots whose only answer is to reinstall Windows.

1 user thanked author for this post.

rc primak

Reply | Quote

At least it’s not Canadian English, eh?

1 user thanked author for this post.

Cybertooth

Reply | Quote

I think they need to develop a version to allow a guy from Boston’s South End to communicate with a guy from Brooklyn, NYC!

-- rc primak

4 users thanked author for this post.

Charlie, wavy, WCHS, Cybertooth

Reply | Quote

If they can do this, why not just have the native’s own language translated and then pushed out as English, then?   Win/win for all.   Amazing they can do this in real time, regardless. Technology’s starting to not just catch up to us, but surpass us.   🙂

Reply | Quote

Probably this is in the works somewhere.

One problem, illustrated in the Snatch scene, is idioms. Words and phrases which have a vernacular meaning not corresponding with a standard dictionary meaning, or for which there is no literal translation, can really vex translation algorithms. Chinese to English and vice-versa is particularly vexing, but Irish to Cockney can be equally diabolic.

-- rc primak

1 user thanked author for this post.

OscarCP

Reply | Quote

I imagine that this is what the algorithms do — translate sounds.

But, it not an easy task.

For one reason, it’s not a matter of converting phonemic segments in the speech of the speaker into the phonemic segments of Standard American English. It gets down the the phonological level. For example, a speaker from the South could be saying “windy” ( but the intent is “Wendy”). Since the speaker’s ‘i’ vowel is before a nasal (‘n’), the translation will convert it to the ‘e’ vowel. But, even then, translating at the phonological level won’t cut it in some instances. The speaker could be saying either ‘winner’ or ‘winter’. They both sound the same and there is no environmental phonological clue to resolve it. It would take some semantic context to determine which it is and to render it in SAE, where the two words are pronounced differently.

I wonder what the algoriths will do with Afro-American Vernacular English (AAVE) “Don’t nobody know” which translates to Standard Amerian English “No one knows” Looking strictly at the AAVE phonology will not get you to the SAE.

And with idioms, if it is only ‘sound’ that is being converted, then “he kicked the bucket” would still be “he kicked the bucket” and not “he died”.

1 user thanked author for this post.

wavy

Reply | Quote

Many native English speakers have a problem understanding anything that is not pronounced within the narrow spectrum of sounds they are familiar with, the one from where they grew up, and idiomatic differences (colloquialisms, ways of saying, grammar, brogue) make things more difficult still (e.g., Southern USA “to do the square” for (I believe … ) “making a turn as if to start to go around the square, when you get to it.”)

I wish the Sanas software, if it ever gets successfully developed, or something like it which works well enough for this particular purpose, were used to make more intelligible the YouTube DIY videos when someone with a strong not-from-here accent explains something like how to get PowerPoint to include tables in slides, let’s say. It is, I am sure, a case where the presenter is making his or her very best to explain something, but I just cannot follow the explanation well enough. However, it follows from the article this discussion is based on, that the Sanas software will be of no help in translating colloquialisms, so those who wish to benefit from its use to communicate with others will need to be careful of how they say what they say.

English is not my first language, so I occasionally have some of those accent comprehension problems with other people understanding me. It is also, in my personal observation, one that tends to become manifest when there is a difference in social class and level of education. But not always: once upon a time, after giving a talk at a meeting in Colorado State-Boulder on a technical approach to monitoring ice melting in Antarctica using GPS, I was asked a question by someone definitely from here, who then did not understand my answer at all, although probably everyone else got it just fine (some hands were raised above smiling faces, clearly asking for  permission to explain, something later confirmed when chatting over coffee during the break) and it was the meeting chair who repeated my answer for his benefit. This is an extreme example that illustrates the point.

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

Reply | Quote

I did not know that real-time speech-to-text translation, I believe the first step of the process you have mentioned (the second being text to speech, something already in use) is that fast, now days, without using special computing hardware not available in the mass market.

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

Reply | Quote

I am a native speaker of English, born in the upper Midwest, where the phonological differences between my dialect and Standard American English (SAE) are very small. (If I had gone to “broadcast school” I would not have to change very many aspects of the sound of my speech. Most changes would be morphological — using SAE “had drunk” or “had ridden” [which we were taught in school to say] rather than the local dialect forms “had drank” or “had rode” [which I commonly heard spoken outside of school]).

I am accustomed to hearing “Broadcast English”. However, in the 3rd example, I found the “American” rendition completely unintelligible after “It was my report from the North …” I would say that Sanas needs to improve its algorithms.

Nevertheless, I applaud Sanas’ efforts and hope that Dell will adopt this software when it comes out. It is only a rare speaker from India that I understand completely. Most of my interaction involves asking an agent to speak more slowly and to “spell” words that I still can’t understand, then repeating back to them what I understand and asking for confirmation. I suppose that the agents understand SAE better than I understand Standard Indian English. I wonder if the algorihms can also convert my SAE to their SIE for them to listen to. Or does it only work one way — SIE to SAE?

2 users thanked author for this post.

wavy, OscarCP

Reply | Quote

So, could this software render Monty Python intelligible to me?  🙂

On another plane altogether, I can picture spy agencies around the world finding great value in this kind of software.

@ScotchJohn already brought up the third one of my thoughts upon hearing of this Sanas project.

 

Keep Running Windows 7 Safely for Years to Come  
Make Windows 10 look and work like Windows 7

1 user thanked author for this post.

wavy

Reply | Quote

Maybe the algorithm would make Monty Python more understandable to people here, but if it did, would it still be funny?

It is a matter of being used to how they talked, that was actually with quite good elocution, perhaps a bit faster than people on this side of the Atlantic are used to hear, and making sometimes funny-sounding voices. This is in line with my opinion of the problem English speakers anywhere have understanding others when the difference in pronunciation deviates more than slightly from what they are used to. It’s quite remarkable.

The problem, I think, is with the spoken language itself: the sounds that are used to pronounce the words. There are more English vowel sounds than there are vowels in the alphabet to write them down, requiring greater accuracy in how they are pronounced, to avoid confusion, unlike in German, Spanish, Italian, or Greek, that are phonetic: what you read out loud just pronouncing the letters (plain or modified in unique ways by the diacritic marks), is how it is pronounced and people can understand it, even if you, who are the speaker saying them, can’t! There are minor regional variations, for example the three consonants that are pronounced differently in different parts of the Spanish-speaking world: the c in “ce” and “ci”, the J and the Y. But this is entirely systematic and takes very little to get used to. Germans, Italians and Greeks, unlike Spanish-speakers, have languages that have not strayed far, becoming a main language of other, distant nations, from the regions where they evolved to their present form. Their variability takes the form of different ways to pronounce some letters and combinations of letters (in German, the different sound of “ch” in different regions, for example) and, of course, also of true regional dialects. One curiosity is Spain, where there are practically no really different regional dialects anymore, but different regional languages, of which “Spanish” is one of them: the Castilian.

So an accent transliteration device (to coin a word), like the one that Sanas aspires to make, would be easier to implement to make speech from one region of the Spanish-speaking world sound as that of another, but actually there would be no much point in doing it. The main difference in speech that sometimes causes difficulty in understanding is the use of colloquialisms, and for that a different type of solution would be needed.

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

Reply | Quote

OscarCP wrote:

The problem, I think, is with the spoken language itself: the sounds that are used to pronounce the words.

Yes.

For those of us whose native tongue is one of the languages that you listed (or some others I can think of), English comes off as phonetically quite chaotic. To illustrate this point, George Bernard Shaw once said that, using characters that can be used in English to represent the desired sounds, one could arguably write the word “fish” as “ghoti.” (Think about it…)

OscarCP wrote:

There are minor regional variations, for example the three consonants that are pronounced differently in different parts of the Spanish-speaking world: the c in “ce” and “ci”, the J and the Y.

Don’t forget about the “ll” as in “llama.”  🙂  My Larousse Spanish-English Dictionary snootily notes that in some parts, “ll often approximates the Spanish y, so that for instance pollo and poyo are indistinguishable.” They neglected to mention that when pronounced the way they prefer, pollo and polio then become indistinguishable. There ain’t no perfect systems.

 

Keep Running Windows 7 Safely for Years to Come  
Make Windows 10 look and work like Windows 7

Reply | Quote

Cybertooth, you are quite right! And “LL” is much more different in its pronunciation in different regions than “J”, because this one is essentially pronounced quite softly or quite harshly, almost like the “G” in Dutch, and that’s it.

And the “LL” is definitely not going to make it easier for the Sanas people to do a version of their algorithm that can be used everywhere.

The “LL”, which is counted as one letter, much as “rr” is also one letter (that sounds like the Scottish rolling “r”), as is “ch” (sounds the same as in English), and “ñ”, that sounds same as “gn” in French, has a very particular sound that is not the same at all among people that hail from three different regions: (a) as “Y”, (b) as “sshh” (but more harshly at the start, pushing the tongue against the front lower teeth and then letting go), and (c) in a way that cannot be described in writing and can only be heard.

So here is an example, although the teacher is not correct in identifying it with the English “Y” (in some places that’s correct, but not in the area I am referring to), as you’ll notice later on, in the sentences given as examples, and it is how I pronounce it myself and as they do in Mexico and in Spain — and for all l  know also in other places, the Philippines, maybe? (Haven’t spoken with a Spanish speaker from there in a long time, so I don’t remember that detail.)

https://www.youtube.com/watch?v=bEd3P-juQuQ

And with this particular pronunciation of “LL”, “pollo” and “polio” are quite distinguishable from each other. As they are with the “ssh” pronunciation from the River Plate region of South America and, I’ve been told, also somewhere in Southern Spain.

By the way, the Welsh also have the “LL”, and they use it in a lot of words, often more than once in the same word, but its sound is a very different one again, somewhat like a short: “ssh” — but not quite.

P.S.: I know that Bernard Shaw, who had a neat proposal for straighten up English spelling that never succeeded, of course, did write that “fish” could also be written as “ghoti”, but in all these years since I read that, I have not figured it out yet. Or maybe there is nothing to figure out?

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

Reply | Quote

Why not use the international phonetic alphabet descriptors for these sounds — See the IPA chart. There’s an audio file for each sound in the chart.

For example, the Welsh ‘ll’ you note seems like it might be the lateral alveolar voiceless fricative. Go to the IPA chart and you can hear it there.

2 users thanked author for this post.

Cybertooth, OscarCP

Reply | Quote

WCHS: That would be a great idea, using the international descriptors, but many among us are unfamiliar with this, so I thought better to leave that out. Besides, I thought the YT video was more fun, particularly since it even had a “bad” word in the background, at one point.

Also, I could not find this particular sound in the IPA list.

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

1 user thanked author for this post.

Cybertooth

Reply | Quote

About that rolling “rr”: Most Spanish speakers pronounce it as in Scotland, but I don’t, because my parents, my parental uncles and and their parents moved, for their own separate reasons, to a difference area from their original one and where both my paternal and maternal lineages, the maternal in particular, have existed and still do for, well, quite a while, long enough to contribute footnotes and short paragraphs to history books. I was teased for the “rr” and the “ll” in school, but I’m persistent (my middle name, practically) and stayed on with that. Also resulting in a few bloody noses, not all mine.

The way it is done by those living in a fairly large but discontinuous territory spread over parts of more than one nation, is like this: one lifts the tongue towards the palate, starts a whistle there (like calling a dog, or stimulating a horse to provide a urine sample for a drug test before a race) and ads a growl, producing a sort of hissing “rhi” sound. Briefly, of course: it’s just a letter, not a speech.

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

Reply | Quote

OscarCP wrote:

I know that Bernard Shaw, who had a neat proposal for straighten up English spelling that never succeeded, of course, did write that “fish” could also be written as “ghoti”, but in all these years since I read that, I have not figured it out yet. Or maybe there is nothing to figure out?

How about, for example,

gh as in enough, o as in women, ti as nation

 

Alice

4 users thanked author for this post.

SueW, wavy, OscarCP, NaNoNyMouse

Reply | Quote

Well, I am trying to give a plain and simple answer to Alice, thanking her and pointing that she is the second person that is kind enough to explain the riddle, and providing some background on where this comes from, but the WebPress filter is completely against it. I have emailed Susan about this most remarkable filter behavior.

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

Reply | Quote

Be nice if they could make an app that parsed lies as dead space or even better rewrote as true speech. Ok no even remotely possible but..

🍻

Just because you don't know where you are going doesn't mean any road will get you there.

Reply | Quote

Wavy, you wrote: “Be nice if they could make an app that parsed lies as dead space or even better rewrote as true speech. Ok no even remotely possible but..”

So perhaps we’ll have to content ourselves with an application like Sanas’ that, instead to Standard American English, transmogrifies their mendacious ramblings into ancient Babylonian. Thus formally preserving freedom of speech and also clearing the air from the metaphorical clouds of dust, ozone and fine particulates they kick up to obscure the sight of the citizenry from the not very nice things that lurk behind such lies and etc.

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

1 user thanked author for this post.

wavy

Reply | Quote

OscarCP wrote:

P.S.: I know that Bernard Shaw, who had a neat proposal for straighten up English spelling that never succeeded, of course, did write that “fish” could also be written as “ghoti”, but in all these years since I read that, I have not figured it out yet. Or maybe there is nothing to figure out?

Heh, heh… That would be the “gh” from the word “tough,” the “o” in “women,” and the “ti” in “nation”.  🙂

My favorite ghoti is salmon.

 

Keep Running Windows 7 Safely for Years to Come  
Make Windows 10 look and work like Windows 7

1 user thanked author for this post.

OscarCP

Reply | Quote

There are serious ethical issues here. Whilst the founders of Sanas AI play the ‘as immigrants ourselves’ card, I question  – what are the accent boundaries? Would Sundar Pichai (Google CEO) need his accent neutralized because…you know… he don’t sound American to me…have you heard him say  “Dev-vel-lop-ment” (just like Apu from Simpsons) – shocking how he got that job.

Will call centers ask employees for permission to neutralize their accents just to appeal to the Americans and Brits who have a problem with foreign accents?? Actually… …they just want to be paid, of course they won’t object.

Anyway, will this tech even work?? It’ll have to accurately transcribe the foreign person’s accent and then give it a whitewash makeover – because we know accents from Harlem won’t cut it.

Even the best transcription technology doesn’t accurately transcribe words from native speakers correctly. Wait…maybe because too many folks training the AI aren’t native speakers. Arrgh…this is complicated. Well, thank God we have courageous entrepreneurs – like the founders of Sanas AI –  who had the balls (founders….all men) to stand up for immigrant issues.

Before I get off my soapbox – here’s a video my Brit friend sent me. Alas, if only Mira used Sanas AI!

https://www.youtube.com/watch?v=YpiAhxaE_oI

 

Reply | Quote