Clearly the best solution is to transmit the data as synthesized voice. Your bitrate will suffer, but it might make it through the other side unscathed.