VoiceTranslator & MouseReader Key Features
VoiceTranslator uses the Microsoft Speech Recognition engine freely offered with Windows XP and Vista. Performance and accuracy was vastly improved in Vista version.
Dragon Naturallyspeaking v9 is not yet directly supported. It can be used in the same way as manual typing into the input text box, and the translation process can be automatically launched whenever the input text box changes (without having to press any key), but since the input text box has to retain the focus for user input, the app cannot recognize sentences with this engine while being on second plane or minimized.
VoiceTranslator makes use of 3 popular translation web services (Google, Babelfish and FreeTranslation)*
Moreover, for offline use VoiceTranslator also includes support for local dictionary text files, that are mainly intended for isolated words, idiomatic expressions and short sentences (the program does not include an true offline "translation engine").
The local dictionaries are easily extendable, by the user itselft and through collaboration with the rest of users. You can add new translations to the local dictionaries in a comfortable way even while you read documents o webpages (via a hotkey), without having to switch back&forth between the app and the document that you are reading. You can select the option to upload your new translations to the global database in order to share them with the community and help to improve the online and local dictionaries.
You can also search translations in the online dictionaries and of course collaborate to improve them, adding new translations, translation tips, example phrases... and voting the ones already entered in the database.
*I've seen some other programs and widgets over there that make use of those web translation services, but maybe they are violating some of their TOS, I don't know. The program tries to avoid abuse (for example, for translation of "speech recognition hypothesis" only permits the use of the local dictionaries) and only intends to provide an improved and more comfortable User Interface to those services. Anyway if any of the previous companies complains about it, maybe I will have to remove them from future distributions of the program. I hope they do not care about it, and maybe any of them hires me instead. :-)
VoiceTranslator and MouseReader use the Text to Speech (TTS) engine freely included with Windows XP and Vista. Besides the engine, you need compatible voices.
Although there are available some free compatible TTS voices over there (mainly from Microsoft and L&H), we advise you to acquire a voice pack from third party vendors like Loquendo, ATT Natural Voices or Nuance Realspeak, which probably will provide a much better sound quality and are also available for many more languages.
Here there is a non exhaustive list of free and comercial TTS voices.
In the long term, maybe the online dictionaries will be the most important part of this project. Their main features are:
- The translations are added by the users (like in the wikipedia example)
- The users can vote the translations, and the ones with the best scores will appear first in the search results (like in the digg example)
- Uses the concept of "translation tips" to help in the selection of the correct translation depending on the context. These tips can also be voted
- Big list of language pairs (maybe they will grow in the future) in the search of the maximum universality of the dictionaries.
SOME TIPS & TRICKS
- In Windows Vista you can install additional language packs to get speech recognition in other languages different than yours (available for US English , UK English, French, German, Spanish, Japanese, Traditional Chinese and Simplified Chinese)
- In Windows XP there are some (old) free speech recognition engines for US English, Japanese and Simplified Chinese. Maybe you will have to install the Microsoft Speech Recognition engine v5.1 if you don't have it. Office 2003 (English version) and Windows XP Tablet Edition included an updated version of the US English engine (v6.1 instead of v5.1) but anyway, speech recognition accuracy is not so good as in Vista.
- I have managed to use VoiceTranslator with Dragon Naturallyspeaking 9 in Windows XP while I work with other applications "simply" by emulating a copy of Windows under VMware and mantaining the VoiceTranslator app in the foreground in the virtualized OS (I know, a very dirty trick, ... I need the Dragon NaturallySpeaking 9 SDK!!!)
- You can use a bluetooth headset for speaking and listening. Currently their sound quality is frequently awful, specially for the microphone input (adversely affecting to the speech recognition accuracy), but they can be more comfortable to wear, and you will not call people's attention as much as if you use common earphones. If you can use them at work, you will not appear so much like being listening to music or chatting.
If you know how I could improve the microphone sound quality using the bluetooth headset profile, please tell me. I've tried even with expensive headsets in different computers but the results are also dissapointing. I think that even bluetooth headsets equipped with the a2dp protocol will not solve the problem, since they do not improve the microphone sound (indeed, I've read that the microphone is frequently deactivated while playing sound). Maybe new headsets with CSR chips and Auristream codec will provide a better sound quality in the near future.
With some headsets and USB bluetooth dongles, there is a time delay between speech activation in the PC, and when the headset really starts to pump out sound correctly. A dirty solution is to mantain the headset constantly activated, for instance by repeteadly playing a .mp3 o .wav file at 0 volume in the background (with Winamp for example). I will try to solve that issue with an option inside the program in the future.
- With MouseReader, remember that in some programs (e.g. Internet Explorer) you can select a whole text paragraph with "triple clicking". Sometimes, depending on text formatting, it can be more comfortable than the manual selection of the paragraphs.
You can use VoiceTranslator as a language learning tool but it also can be useful in some other situations:
- You also can use it as a speech recognition aid.
For instance, it can help you while attending live or online international speeches, conferences or meetings.
Although user training vastly improves speech recognition, both Microsoft Speech and Dragon Naturallyspeaking 9 allow you to create new users without specific training and speech adaption, so you can get a somehow "speaker independant" speech recognition tool, and obtain rather good results if the speech is clear and the background noise is low. I remind you that a correct selection of the UK or US English accent can improve substantially the recognition accuracy in English.
Maybe it is a subjective perception, but (to me) it seems that creating new users for each new speaker yields better results (you can delete them later). Maybe the recognition engine makes some kind of adaption even without specific preliminary training.
- Besides the usual audio input from the microphone, many audio cards have the option of setting as input source for recording the "Stereo Mix" (or somewhat similar in your system). This is interesting as you will be able to pipe sound played on your PC directly to the speech recognition engine, and recognize speech from local or online videos, TV, radio, podcasts, movies, audiobooks, videoconferencing with skype, messenger...
Each time the VoiceTranslator app is closed and re-opened, the input source will return automatically to the default input device set for speech recognition in your system (usually the microphone). If you want to use a different input (Stereo Mix, for instance), you will have to set it as the default input device for speech recognition, under Control Panel (or selecting it again when you re-open the app). The same applies to the Text to Speech system default voice. Probably another thing to improve in future versions of the programs.
If your sound card does not support the Stereo Mix option, you still can "close the loop" using the PC speakers and a microphone (if you are in a quiet room) but this intermediate step usually will worsen a lot the speech recognition accuracy. If your PC has a LINE-IN input jack, in principle you may close the loop connecting the headphone output to the LINE-IN input with a proper audio cable (Caution, NOT to the microphone input, its signal level is usually much lower) and then setting the LINE-IN as the input device for speech recognition. The problem is that then you cannot hear the audio, so probably that will be useless apart from testing purposes, unless you use 2 computers, or a proper Y bifurcation audio cable (like the ones used to share a mp3 player)
- VoiceTranslator has an option called "Pronunciation Feedback". When it is activated, recognized text is played back through the currently selected Text To Speech voice, so you can polish your pronunciation in a foreign language this way (of course, that is supposing that the program has recognized correctly what you said, even if pronunciation wasn't perfect. If not, maybe you will learn new words :-)
- VoiceTranslator can also be useful for online gaming, running in second plane, for recognizing other players speech and also for helping you to answer to the typical self-question of non English native speakers: "How can I say in English...?"
- As the dictionaries content is generated by the users, currently only the Spanish-English and English-Spanish local dictionaries have "some" meaningful content. I hope we can further improve this and the rest of dictionaries with the collaboration of the users of the desktop app and the online dictionaries.
You can access the online dictionaries via voice-translator.com/searchtranslation.php and also via the short domain googl9.com that is shorter to type and maybe easier to remember. I registered that domain for another project that I haven't finished yet. I have not registered a proper 2.0 web name, such a transvoizr.com o voiztranz.com, and maybe I will not become rich because of that. :-)
- Unicode compatibility is not fully implemented and tested, so it is very likely that the program does not work correctly with some languages with special character sets.
- It is needless to say that neither Text to Speech pronunciation, nor speech recognition and even less the translation process will work always with 100% accuracy.
But I think that "something" is better than "nothing", even further when that "something" is constantly improving and approaching more and more the right answer (I'd wish I could speak and understand English as well as my computer does.)
You can read more tips and tricks here.
A demo of VoiceTranslator can be downloaded here.
A demo of MouseReader can be downloaded here.