Kõnele is an app that helps other apps to communicate with two online speech recognition servers, running the following software:
- continuous full-duplex server, https://github.com/alumae/kaldi-gstreamer-server,
- grammar-supporting server, https://github.com/alumae/ruby-pocketsphinx-server.
You can benefit from Kõnele either by implementing a new app and using an existing running server, or deploying a new server and using it from an existing app.
Calling Kõnele as an activity
Kõnele implements android.speech.RecognizerIntent actions
and supports its EXTRAs up to Android API level 3.
In addition to the standard EXTRAs, Kõnele adds the following EXTRAs:
||URL||Web address of the speech recognition server|
||URL||Web address of a speech recognition grammar file|
||Comma-separated language codes||One or more identifiers of languages into which the recognition server should translate the raw speech recognition output|
||String||Desired transcription (could be used for adaptation)|
||Boolean||Return audio iff true|
||Mime type (only “audio/wav”)||Audio format|
GET_AUDIO EXTRAs are prefixed by
android.speech.extra, all others by
The full list of such additional EXTRAs (some of them only experimentally supported) is part of the speechutils-project.
Note that the end-user can override the server and grammar EXTRAs via the Kõnele settings, i.e. your app should not assume that the specified server or grammar was actually used.
If you know that Kõnele is available on the device and is the only one with the required features
then you can call it directly, i.e. without any intermediate user-selection.
To do this, build a Recognizer-intent that can only be serviced by Kõnele’s
Calling Kõnele as a service
You can also call Kõnele via android.speech.SpeechRecognizer.
In this case, please use
EXTRA_CALLING_PACKAGE to identify your app for Kõnele.
To obtain a Kõnele-specific SpeechRecognizer-object, use the two-argument call to
The available services are:
ee.ioc.phon.android.speak.service.WebSocketRecognitionService(recommended; uses the continuous full-duplex server)
ee.ioc.phon.android.speak.service.HttpRecognitionService(uses the grammar-supporting server)
The above-listed EXTRAs are also supported when calling Kõnele as a service, with the
exception of the
Calling Kõnele via Android Debug Bridge (adb)
The following two
adb shell commands
- stop a running instance of Kõnele (if present)
- start Kõnele in a mode where the expected input is an action command in English (e.g. “one plus two”) and the output is the corresponding expression (“1+2”), which gets opened in a web-browser
Kõnele calls its own services via the standard interfaces
and classes which contain “Demo” in their names contain sample code.
Here are links to open-source apps that use Kõnele: