Geraint and Gwyneth speak out

At last year's Eisteddfod, it was announced that a new synthetic voice capable of reading Welsh text and converting it to speech had been developed by a Polish company called Ivona. However it was at that stage only a demonstration rather than a fully working product. I've checked a few times since then to see if it had been developed further, and I'm pleased to say that it has. There is now a male and a female voice, and as an example of how they sound I've taken the text from a story on the BBC's Newyddion site and recorded it in each voice:

     

     

Ysgol newydd: Pryder am y Gymraeg yn Nyffryn Tywi

Dywed rhieni yn Nyffryn Tywi na fydd ysgol uwchradd newydd yn ardal Ffairfach, Llandeilo, yn bodloni'r galw am addysg cyfrwng Cymraeg. Bydd yr ysgol newydd yn cael ei hagor yn 2015, a bydd ysgolion Tregib (Llandeilo) a Phantycelyn (Llanymddyfri) yn cau.

Dywed rhai rhieni fod angen ysgol sy'n dysgu'r rhan fwyaf o bynciau, ac eithrio gwyddoniaeth, drwy gyfrwng y Gymraeg. Ond statws 2B fydd yr ysgol newydd, lle mae yna ddewis i dderbyn yr holl wersi drwy gyfrwng y Saesneg, neu 80% drwy gyfrwng y Gymraeg.

Yn ôl rhai rhieni dylid sefydlu ysgol categori 2A, fel Ysgol Maes yr Yrfa, lle mae 80% o'r pynciau yn cael eu dysgu drwy'r Gymraeg. Bydd yn rhaid i deuluoedd sydd am i'w plant gael addysg 2A anfon eu plant i Ysgol Maes yr Yrfa yng Nghefneithin.

Dywed Dr Hywel Glyn Lewis, darlithydd mewn dwyieithrwydd, y dylai Cyngor Sir Caerfyrddin ystyried gweithio ar y cyd gyda Chyngor Powys a sefydlu ysgol 2A ar gyfer yr ardal. Byddai hynny o bosibl yn golygu gwasanaethu ardaloedd Pontsenni a Thecastell yn ne Powys. Pryder Dr Lewis yw y byddai plant sy'n mynychu ysgolion cynradd cyfrwng Cymraeg yn colli'r iaith drwy fynychu ysgolion uwchradd lle nad oes darpariaeth ddigonol.

Mae rhieni ardal Llanymddyfri hefyd yn anhapus fod yr ysgol newydd wedi ei lleoli yn ardal Llandeilo. Dywed ymgyrchwyr y dylid wedi dewis ardal sydd rhwng Llandeilo a Llandysul.

BBC, 20 July 2011

OK, it's not absolutely perfect, but it's miles ahead of previous attempts to get a natural sounding synthetic Welsh voice. If you want a comparison, try using the "Darllenwch y dudalen" feature on say this page of the Bwrdd yr Iaith website. Highlight some text and click "darllen testun".

If anyone wants to try out the Ivona voices, just put some text into the box on this page, then choose either Geraint or Gwyneth from the drop down menu to read it out. They'll even read things in English, but with a Welsh accent. That's cool.

The voices are still beta versions and not yet for sale, though I expect they soon will be, maybe in time for the Eisteddfod. When they are, the personal version of the programme should be able to read Welsh text from any computer application or web page. But I hope we will also see the BBC, Golwg360, councils and private companies install a button which will read the text on their Welsh pages automatically.

Bookmark and Share

6 comments:

Siônnyn said...

It's fantastic! Had I heard this casually, I would not have realised that it is computer generated.

How do they do it? The intonation is (mostly) especially convincing.

Cymro i'r Carn said...

Mae ansawdd yr ynganiad hefyd yn wych ware teg. Mae hyn yn gam mawr ymlaen, Lot o ddiddordeb yn y Gymraeg yng Ngwlad Pwyl yn ddiweddar, sgwni os taw mewnfudwyr sydd wedi symud yn ôl o Gymru sydd wedi sbarduno'r ddiddordeb.

MH said...

There's a bit of information on how they to it on the Ivona website, Glyn. The voices are derived from recordings of a human voice, but all the skill is in stitching those fragments of recorded sound together. I think the key is not to concentrate on just the words (which is what older programmes did) but on how words flow into each other in natural speech.

I think the spur to doing it has come from the RNIB, because the programme is obviously of greatest benfit to blind and partially sighted people. I think the RNIB will make it available for free to them. It will also be very useful for those with dyslexia. But it is quite something for Welsh to be one of the early languages to be developed, alongside Polish, English and Romanian! French, German and Spanish are only just becoming available. Perhaps it does have something to do with the relatively high number of Polish immigrants we had/have in Wales, Cymro i'r Carn. Certainly living in Wales for a time will have shown them that Wales is quite different from England, that the UK is not monolithically English (as is the preception in many parts of the world) and therefore raised our profile as a nation.

But there might be more practical reasons too. Welsh is probably one of the easiest languages to get right because the spelling nearly always matches the sound and the stress is nearly always on the last but one syllable.

-

I've been giving some thought about how useful it will be as a tool for learning Welsh and developing the use of Welsh. I don't think it's quite good enough for beginners learning to speak the language. But it will be good for those who can speak the language or understand the spoken language (because they will be able to "filter out" the words or intonations that aren't quite right) in the sense of increasing their reading skills.

For example, there are older and middle aged people who can speak the language perfectly well but find it difficult to read the language because because they did very little Welsh when at school. They will benefit from this as a learning tool. The same will be true of those who learn Welsh primarily through hearing and speaking, as in the SaySomethingInWelsh course. Once they attain a degree of fluency (though it would be conterproductive to do it too early) they can use this as a way of adding reading to what they can understand and speak.

From the other direction, there are some people who have good reading skills, but not very much confidence in speaking. This particularly applies to more academically inclined people and those who live outside Wales or in areas of Wales where there's little opportunity to speak Welsh. Hearing the sound of a written sentence they can understand but would have difficulty in reading out loud should help develop their conversational skills.

This is a big step forward, as Cymro i'r Carn says, and with a bit of imagination we can use this tool in ways far beyond just helping those with sight problems.

Dafydd Tomos said...

Many organisations have produced Welsh TTS voices over the years. Some were released as open source, others are closed but web-based products, others are standalone software products that read all text on a computer screen including emails and web pages.

The Ivona voices are certainly the highest quality voices created so far but that doesn't mean they can be easily or widely adopted as they are commercial products.

The best solution is for blind and partially sighted people to be provided with this software on an individual basis, which is the intention.

I'm dubious about the general use of 'overlaid' TTS engines on websites for a number of reasons. They can compromise the integrity of the site design (and may break functionality). They depend on third party services being available (and paid for by someone). This is the approach used by the Welsh Language Board for example (they pay for a service from Readspeaker which inserts a poorly integrated TTS element into the website).

This is a novelty but is generally not particularly helpful for blind people who are better served by applications that translate their entire interface to speech rather than particular sites or software that support it.

Often, adding an overlayed speech engine on websites is a quick get-out clause for 'accessibility' purposes, but it's a poor solution.

There are plenty of other useful applications that TTS can be used for where a speech engine is run on the web server and its output is integrated into the site. This type of use will depend on the costs for that server-side software, its requirements and the practicalities of integrating.

Of the products that Ivona offer, it's the 'Speech Cloud' offering that is most exciting as long as the cost of using it can be supported.

MH said...

Sorry your comment has only just appeared, Dafydd. It had been caught in the spam filter.

MH said...

I'd agree with pretty much all of what you've said, Dafydd. For people who are blind, a much more radical solution that incorporates the whole interface is required.

But the important thing is that these good-quality Welsh voices now exist, and programme and website developers should be able to find new ways of using them. It is (or will be) a commercial product, but I found this thread on ForumWales yesterday. It's dated March 2009, and seems to mark the beginning of the project.

We didn't choose the Welsh language as our next TTS [text-to-speech] language for commercial reasons. This is part of an initiative we have undertaken to build, once a year, a TTS engine of languages that are in some way threatened. This is our first attempt in this direction, as we understand that there is a constant decline in the numbers of Welsh speakers (even though the introduction of the Welsh Language Act is slowing this process a bit).

I can only thank Ivona for their attitude, although I guess they would have got some funding from the RNIB and maybe BYIG to help. But it suggests that Ivona might be open to letting certain organizations or groups use the Welsh voices on more favourable terms. So if people have good ideas, get in touch with them. The voices are SAPI 5, which means they will integrate with Microsoft operating systems and programmes; and they also work with Android.

Post a Comment