voice data set

we are looking for Audio data set (Telephonic conversations) that can be used to train speech to text engine for speech recognition. For this we need variety of speech data for Modern Hindi (Hinglish) collected from a varied set of speakers equally distributed across age and gender. Each audio file should focus on one primary domain and on one primary topic with Dual-channel recording on 8kHz sampling rate. Style of Speech should be conversational, unscripted, between two speakers and around 20 to 60 minute long in each audio file. Alongwith Modern Hindi (Hinglish) we would require around speech data for each Bengali, Kannada and Marathi as well.