Automat Parsing of Audio Recordings. Testing Children with Dyslalia. Theoretical Background

In this paper we present our researches regarding automat parsing of audio recordings. These recordings are obtained from children with dyslalia and are necessary for an accurate identification of speech problems. We develop the ADM algorithm and we analyze the complexity of this solution. The main objectives of this task are:  recording 120 children (60 with correct pronunciation and 60 with dyslalia)  we must permit different audio environments during recording (some phonemes will be used for training a real time recognition system)  the cost of recording devices and the children’s impact must be minimized  after recording is necessary to split the stream into phonemes  the speech therapist’s voice must be ignored We utilize a digital voice recorder in High Quality mode and with VCVA (Variable Control Voice Actuator) activated. The record format is IMA-ADPCM, 16KHz and 4bits (16 bits PCM). A microphone was placed at 10 cm from mouth in order to minimize environment noise. A software set of classes (C#) was created for handling audio stream (read, conversion between different format, write). We also propose an original solution for placing markers in audio stream. These markers are needed for a correct parsing af full recoding.