I'm currently working on a project that makes use of the Azure Speech API and I've seen the current documentation on language support. My two questions are...
Is there is a road-map somewhere for future additions to the language support list? Specifically, will there be an standard voice added for Urdu?
And also, for our South Asian friends. I know Urdu and Hindi are different written; but if I understand correctly, they are similar spoken. Are they similar enough to listen for Hindi (supported by Speech-to-text) for both?
Related
Both have their own documentation and I see only small wording differences between those. Are there list of things that have actually changed? Has OCR for example improved on version 2.0 or it's the same except I guess the handwriting recognition? Some kind of changelog would really make a difference.
The only difference between v1.0 and v2.0 is the revised /recognizedText which has a breaking change in the input/output. All other endpoints are exactly the same. Also, if you have an key in an up-to-date pricing tier (including free), your API key will work in both versions.
As you may know, the Computer Vision API has two different OCR endpoints. The /ocr endpoint runs the older recognition engine with broader language coverage. The newer /recognizeText endpoint, which in v1.0 handled handwritten text, in v2.0 covers both handwritten and printed text using a newer engine. The /recognizeText endpoint remains English-only for now. You select between handwritten/printed modalities using the mode query parameter. See
documentation here.
As for documenting changes, there isn't one obvious place for this unfortunately. One option is to check the swagger repo.
Google started “Google Neural Machine Translation” system in 2016 and improved the efficiency of translation of some languages tremendously. These languages include German, French, Spanish, Portuguese, Chinese, Japanese, Turkish and Korean; as they said.
However Dialogflow supports all of these languages except Turkish. Is there any specific reason for this exclusion?
Dialogflow does not handle translation, so Google's work in this area doesn't apply.
What Dialogflow does do that is language related is let you create sample phrases, in that language, and can handle other, similar, phrases that a user might make in conversation. So it isn't about translating a language - it is about natural language parsing and understanding.
I want to write a app (initially Windows) that include handwriting to text recognition. I want to use the Windows built-in Tablet PC INput. My question is is there a way to capture the strokes as an image, "send these to the OCR engine used by the Tablet Input, and return the recognised text?
Or, are there any good open source handwriting libraries that could be used directly?
The primary development language is Qt.
I am not aware of any open source or free software libraries for handwriting recognition, so I wrote an adapter. My target was my tablet PC running Linux, but part of my solution can also be used directly on Windows, although you will need to adapt it to your needs.
You will need to read through the licenses for the components I used and validate your own use of them.
The source is available here: Ink2Text project
Part of this solution is a server which uses the XP Handwriting Recognition libraries to interpret the strokes which make up handwriting. As an aside, this does not use OCR - it uses connected graphs of the flow of the strokes.
Another complementary project provides a client handwriting widget: Stylus/Handwriting Input Panel. This is written in Java, and it's GPL3. It accepts the handwriting and sends it off to the server. Unless you wish to use it as is, it's of value solely to see the data format for the ink, although that's simple enough and you can probably deduce that with just the Ink2Text source code.
An earlier solution used the S/HIP with my MS Ink Server, which accepted input over regular network connections. That may also be useful depending on your architecture, but requires a running copy of Windows.
This system provides very good recognition of printed and cursive handwriting.
I will answer questions about it only in it's associated SourceForge forums, so that others may benefit from the answers as well - please don't ask here.
Cheers,
Bret
I want to be wrong, but unfortunately, there is no available open-source offline handwriting recognition system even close to MS' or Apple's Ink.
On Windows you can play with Ink Recognition (About Handwriting Recognition, Advanced Recognition Sample). C++ interface is available, but not as well documented, as .net implementation is. So, you need to apply more efforts and do a lot of research to achieve what you want.
For another systems (including Windows too) there is way to use Tesseract-OCR with your application. See Tesseract's base api. For better recognition quality, you may train tesseract and use your own trained data.
If you do not want to spend your time doing R&D tasks above, you can use paid solutions like: MyScript SDK, WritePad SDK and so on...
Is there any free dictionary I can use for i18n?
Free as in open source / creative commons, ideally also for use in a commercial product.
Looking at the KDE i18n projects, they have translated a lot of applications in many languages. Is there a way I can use their dictionaries for a standard Qt (non-KDE) application - and am I allowed to?
You should contact the KDE localization team if you have questions about licensing of their translations.
I don't think that the l10n support of KDE applications will help yoiu directly -- they ship as a catalogue of strings, as appears in a particular context in the original application, and the translated form. There is a long way from that to automatically using the data in the context of another application, and that's also the reason why machine-generated translations have such a low quality. If you cannot speak a language and don't have anyone who could do the work for you, you won't be able to ship a working localized version in that language.
Is there an in-browser, Flash-free method to view ebooks from Adobe Content Server?
We currently offer a library of c.50k specialist ebooks via a feature-rich "ereader" web-app (HTML5/JS based with various fallbacks down to IE7).
However, management want to be able to offer "downloadable" ebooks for "mobile devices". By this they mean a file that the user can download and read offline. Adobe Content Server is fine for this (if a little expensive, and a little hated by the users, but unfortunately it's becoming an industry standard...)
OK so if we adopt ACS, making downloadable-for-offline-reading a possibility, what are the options for online reading, assuming we want to use ACS for everything and not just offline? In other words ... is there an in-browser reader for Adobe Content Server?
Flash is not a possibility as a) a lot of the users us iPads (yes for online reading too) and b) a lot of the users have to use IE7 with no Flash installed (the NHS is a major customer).
I realise I might be asking for the impossible but I thought it would be worth hearing peoples' thoughts.
Please don't advise me not to use DRM, it's not my choice and I have already advised against using it. However we are contractually obliged by our suppliers to have "a DRM solution" for offline reading.
If there was a widely-available alternative solution to Adobe Content Server I'd be interested to hear about it. I have already integrated ACS once (version 3) and don't really look forward to repeat the experience...
There are a number of reader apps which support ACS, such as Sony Reader. Your readers can use those apps (after "sideloading" your books, a process which differs from reader to reader) to read the ACS books. I don't know of any browser-based reader, but it seems to me that the apps (which exist for all major platforms) should get the job done for you. These apps all keep local copies of books and work just fine offline.