Nowadays there is an important increasing interest to obtain transcriptions from multimedia repositories and digital libraries. Transcriptions produced by human transcribers can provide high quality results. Nevertheless, the overall process is very slow and usually expensive. One way to deal with this important drawback is to produce automatically transcriptions based on automatic systems. However, automatic transcriptions are still far from producing the desired quality out of some specific scenarios. Indeed, an important human effort to supervise the automatically produced transcriptions is mandatory. Supervision effort can be reduced by means of the use of interactive systems. In interactive systems a human operator is placed at the center of the transcription process and embeds an automatic system within an interactive editing environment. The automatic system and the human transcriber tightly cooperate to generate the final transcription, thereby combining human accuracy with system efficiency. Confidence estimation (CE) has been largely applied in fully automatic systems to predict the output reliability. Nevertheless, its use to interactive systems have been slightly explored. CE could be used to improve performance of interactive systems in different ways. On the one hand, supervision effort should be reduced if only low-confident output parts are supervised by the user. On the other hand, better automatic transcriptions should be produced improving the underlying system models based on supervised and high-confident output parts. The main aim of the project will be to explore these novel strategies in interactive speech and handwritten transcription.