This page in Russian
Current version: 2.7.8
The most prominent feature of this synthesizer is its capability to distinguish and switch language automatically according to the nature of input text. In fact, not all supported languages can be recognized automatically, but only a certain subset. For the moment it includes Arabic, Bulgarian, Chinese, Croatian, English, French, German, Hebrew, Italian, Japanese, Polish, Portuguese, Russian, Spanish and Turkish. Besides that, this synthesizer allows to adjust voices separately. So, such parameters as volume, speech rate and voice pitch can be tuned for each voice individually.
Another important feature of SmartVoice is its capability to recognize and read emojis and it is possible to assign a separate voice for it.
Current speech language and voice can be as well immediately changed manually at any moment via the notification panel widget which in turn can be enabled or disabled in the general application settings.
After installation the synthesizer becomes available in the applications list that allows to setup it just before starting.
Application interface is available in English, Italian, Spanish, Croatian, Polish, Hebrew and Russian.
By the way, I'd like to thank Adi Kushnir for Hebrew and Zvonimir Stanecic for Croatian and Polish translations.
For the most part this product is intended for the visually impaired people who are forced to use speech feedback as a primary interface. Android accessibility service in its current state does not provide any convenient way to change speech language on the fly without additional disturbances, therefore, visually impaired users are now totally frustrated and discriminated in multilingual environment. Moreover, some synthesizers require voice installation as a separate step before anything can actually be spoken.
At the contrary, SmartVoice can speak just after installation without any additional step, because it has its own embedded voice, though only Russian for the moment. For other languages SmartVoice can use Vocalizer Expressive voices from Nuance Communications, Inc., but it is not supported or even approved in any way by this company and, of course, does not include these voices itself. I suggest that the voices should be purchased somehow from the official owner. Anyway, it is not my business. The only intention of mine is to provide so desperately desired functionality to whom it may concern.
So, grab the package here if you like, install it and use at your own risk. Of course, no warranties.
Since the section of application settings devoted to automatic language recognition contains a plethora of options that might look not so evident at the first glance, it seems to make sense giving some explanation about how this mechanism actually works.
At first, all this machinery can be totally disabled by the option "Use only default voice or the one requested by client". If it is checked, SmartVoice will obey to a client request and use only the language and voice specified, just like other speech synthesizers usually do.
In general, text fragment processing in SmartVoice depends on the option "Use one voice for entire message". If it is checked, each text fragment is treated atomically. SmartVoice only tries to guess the voice mostly fitting to pronounce it. In the opposite case some lingual markup is undertaken, i.e. the text is split onto the monolingual fragments and an appropriate voice is chosen for each of them separately.
Since reliable and unambiguous language recognition for rather small chunk of text is somewhat problematic, the priority list specified by the option "Language detection order" comes into view. This list is traversed downwards from top to bottom, thus, the languages placed upper in the list have more chances to be used. It is also taken in account in the lingual markup process.
Such languages as Arabic and Hebrew use their own character sets (alphabets) by wich they can be easily recognized. The others are subdivided into groups by character set community. SmartVoice distinguishes three such groups: Latinic, Cyrillic and CJK ideography. It is possible to choose fallback voice for each of them that will be used when no particular language has been detected, but only the alphabet group.
In summary, the voice for an entire message is chosen as follows:
The lingual text markup process consists of four stages:
Perhaps, it is worth yet to add that only general languages can take part in the recognition process, but not their variants or dialects. But the latter ones can be used in system locales and client requests. For instance, when English language is detected the voice defined for the mere English is used, but a client can request British or American English and in that case the corresponding voice can be used. This notion should be kept in mind when editing preferences in the "Voices assignment" section.
Note about explicit voice switching via the notification panel widget. It acts not like a persistent setting in the sense that the choice only lasts until one of the following events is encountered:
Each list item in this widget accepts long press. For a voice item it opens the respective voice settings screen. Long press on the last item opens automatic language recognition preferences section.