I am trying to use the paid version of google-cloud-speech speech-to-text feature with "Detecting language spoken automatically" feature. The problem is if I give 2 alternate languages (in my case "en-IN", "hi-IN"), then it is detecting whole speech as Hindi only though I have spoken initial 20 sec in Hindi and next in English. I don't think this is the expected behavior so please help me out resolving this issue.
Thanks in advance.
Related
I need to extract people names from documents in German (not my native). After a bit of search, I've found GATE framework which seems to support English, German and many other languages. The accuracy for English is quite decent, but it's unacceptable for German (see screenshots).
Here are the PRs:
And a chunk of highlighted people names:
A friend of mine says that none of these is a person name, so I wonder if I misconfigured something. Do I need to specify the language somehow?
Solution: Install the Language: German plugin and use it. The accuracy is still poor though, at least for my case.
I'm looking for an open source library to detect the spoken language used in an audio file, such as a wav file.
I tried CMU sphinx , but was not able to understand how to use it for language detection? Can someone please help?
If you are willing to learn another toolkit, you should consider Kaldi [1]. It is an open-source speech recognition toolkit with a speaker recognition system (which uses similar models as a language identification system) in the trunk and an experimental language Identification setup in the sandbox language_id. After checking out the repository, you can switch to the LID sandbox with svn switch ^/sandbox/language_id. The LID examples are in egs/lre07.
Whichever toolkit you use, I recommend an i-Vector based system instead of a phonotactic system. An i-Vector based system will be easier to setup as it doesn't require transcripts, and significantly faster, since it avoids decoding.
You can try CMU sphinx in all phone decode mode.
Train models for languages you wish to identify.
Pick language for which hypothesis score is best
I'd like to create a TTS system for a native american language (wayuunaiki).
The language is written in latin (western) alphabet.
I also have information about the phonetics (the rules to convert each word into IPA symbols).
I'm planning to create a database of voice recordings from the native people. Then I want to somehow train that data, using the IPA equivalency information to generate a more accurate speech model.
I'm totally new to Natural Language Processing, so my question is.. which tools can I use to perform what I'm planning?
I've heard that HTK ans CMU Sphinx are quite good in speech recognition. No idea about speech generation. Also heard about Festival, but i read it only uses predefined most known languages: English, Spanish, and so.
Excuse my typing faults. I'm still learning English. Thanks in advance!
You can add new language in Festival, it's actually specifically designed to simplify new language creation. For more details read the festvox book:
http://festvox.org/bsv/
Another toolkit to consider is OpenMary, see their documentation too
https://github.com/marytts/marytts/wiki/New-Language-Support
It is more modern and might be easier for you.
In any case you will have to spend some time and write the code to describe your language. Usually it's about 300 lines of code. After that you can record single-speaker TTS database and run voice building process. The more you record the better the result would be.
Use Festival toolkit for text to speech (Tips : Use Linux operating system)
For production of a set of mobilephone/ smartphone minisites, what do you recommend as a technology to automatically choose the language of the site:
browser IP address
mobile browser language request header
any method related to device specifics or Carrier specifics of a certain country?
any other method
The languages that will be targeted are:
Vietnamese
German
Thai
Arabic
Spanish
Indonesian
Italian
Japanese
Chinese, both traditional/ Simplified
Korean
Russian
I understand the answers may vary per language, so feedback on all or any language would be greatly appreciated.
Anything other that what a user has specifically requested is a bad idea. So, for example, using geographical IP lookup is a terrible idea. People may live in a country where multiple languages are spoken, or may simply prefer the lingua franca English, and might find it extremely annoying when another language is forced upon them.
Out of the options you mentioned only the browser language request header sounds like something a user might actually configure on his own. All other options I suspect will produce an inferior experience for large portions of the target audience.
I'm currently working with a dead network application that received packets for a chat.
I checked that it received text like this one:
hi there! {c:0000FF}foo{/c} sentence
I have checked but have not found the language that uses that color syntax. It is some famous language or it's surely a self made script/library?
Original application used c++ and python as developping languages.
Thanks you all in advance,
Rag.
Hard to prove a negative, but I'll wager it's a proprietary ('self made') syntax.