I cannot figure out how to register an unregistered account originally published.
In any case, I'm not interested in the audio format right now, but just the streaming aspect. I want to take the microphone input and transfer it from the iphone to the server. Currently, I do not care about the transmission speed, since I initially tested the Wi-Fi connection, and not the 3G setting. the reason i can't cache it is because they are interested in trying some open source speech recognition materials for my graduation thesis. caching and then sending a recording is possible, but then it takes much longer to get voice data to the server. if I can start sending data right after the start of recording, the response time will improve significantly, because most of the data will already reach the server by the time I release the record button. in addition, if I can get this streaming function to work with iphone, then on the server side I can also start the speech recognizer as soon as the first bit of sound starts. again, this should significantly reflect the final amount of time that the transaction takes from the user's point of view.
colin barrett mentions telephones and telephone networks, but in reality it is a rather suboptimal solution for asr, mainly because they do not provide a good way to recover from errors - this is due to the voip dialogue - this is a terrible experience. however, iphone and, in particular, the touch screen provide a great way to do this using ime or nbest lists for other candidate candidates.
if I can define a basic architecture for audio streaming, then I can start thinking about flac coding or about reducing the required bit rate. it is even possible to extract traits, although this limits the later possibility of retraining the system using records.