A Framework for VoIP Speech Data Generation Using Asterisk

State-of-the-art approaches for automatic recognition of speech, speaker or language specific information from spoken data rely on statistical techniques that require large databases for training and testing. Application of these techniques on Voice over Internet Protocol (VoIP) environment requires studying them under different codec and network conditions. Though earlier works have studied and reported the same, a framework for automatic generation of VoIP speech is lacking. A number of speech corpus for different applications are available for microphone speech. As domain specific performance needs to be evaluated in matched acoustic characteristics and application conditions, methods that enable automatic generation of target speech from the available microphone speech are important to a researcher in saving time and effort. We present a framework based on Asterisk, a freely available open source Internet Protocol-Public Exchange (IP-PBX) software for realization of VoIP speech from the available microphone speech corpus in network conditions that is reflective of actual VoIP channels.