libWhisper library is intended to provide specific application level
objects, generaly based upon libWhisperCore classes. Obviously, this
allows applications to share a common core.
Basicaly, a data handler is a detached thread that can have an incoming data stream and one (or more) outgoing data stream(s).
Please refer to libWhisperCore for more informations about
data handlers (rtDataHandler) and data
streams (rtDataStream).
This library provides, among other things, non-abstract useable data handlers that could be used in any application using WHISPER.
rtMicrophoneHandler, inherited from rtDataHandler, is the main
class for microphone handling. It is the only one that has to be instancied
by programmers. Its work is to initialise the microphone, and codes incoming
audio streams into the outgoing buffer of data handler.
Audio encoding is made by an instance of rtMicrophoneCoder that
belongs to a rtMicrophoneHandler.
Please note that rtDataHandler standard block handling method is
ignored and that the execute(void*) method is overloaded (since
no incoming stream rtDataStream exists).
First of all, rtMicrophoneCoder is doing a basic microphone
calibration. When done, it enters in a normal audio encoding mode that
consists in listening for noise, ignore silences except when they reflect
a end of speech stream (EOS), and chain either sound blocks
(ST_SOUND data stream block type) or EOS block
(ST_EOS type).
The silenceBlocksForEOS parameter contains the number of silences
corresponding to the end of speech/stream information.
Others parameters handle microphone hard configuration :
acqFrequency (in Hz, default 8000), sampleSize
(is the basic word's number of bits), blockSize
(depending on underlaying hardware, this is the byte number that are
readen from device in a raw; commonly 64), and blockNumber
is the number of raw device blocks that'll be in encoded a
sound block buffer.
rtSpeechRecoder class can be considered has a caching-only
data handler. Is simply recodes incoming sound blocks into its outgoing
buffer, using the handleSound(rtDataStreamBlock *) method
of rtDataHandler.
This class inherits all standards parameters of rtDataHandler,
and use the normal stream handling method. Refers to this class for
more explanation on the parameters and behaviour.
Phones handler is providing a continuous listening of incoming audio
blocks. To do that, it uses the EOS information, originaly
given by rtMicrophoneHandler (which can be a remote handler).
Since rtPhonesHandler is derived from rtStreamWriterFile,
it uses an overloading of the write(rtDataStream *) method that
is used by parent's handleSound(rtDataStreamBlock *).
Silences and parameter changing into previous handlers are ignored
(waitingBlocks is simply decremented in
handleSilence(rtDataStreamBlock *) and
handleParameters(rtDataStreamBlock *) methods.
The rtDataHandler readConditionOnLast(rtDataStreamBlock *)
now waits for an EOS block type or waiting blocks to be above
10 (default value). This affects the rtDataHandler behaviour.
rtZLibCompressor handler compress incoming audio stream into
ST_ZDATA blocks using the zlib compression library. Thus, the outgoing
stream should hold compressed audio blocks (and possibly some other
block types).
rtZLibDecompressor does the opposite : it decompresses each block
of type ST_ZDATA into ST_AUDIO block.