Gravar e-mail: New experiments on speaker diarization for unsupervised speaking style voice building for speech synthesis