Diacritics Restorer v2

The Diacritics Restorer v2 web application performs diacritics restoration for fully-formatted .docx files written in Romanian and was developed as an improvement of an existing web service (Diacritics Restorer). The key feature of this application is that it maintains the formatting for titles, paragraphs, tables, captions etc. The application allows the user to upload the desired .docx document, restores the diacritics and sends the final .docx file to the e-mail address specified by the user. The user can also download directly the final document using a single button which becomes visible after the restoration ends.

Live Transcriber 2017

Live Transcriber 2017 is, to the best of our knowledge, the first DNN-based large vocabulary automatic speech recognition system for Romanian language. It transcribes your speech using a vocabulary of 200.000 Romanian words, with an average accuracy of over 90%. The acoustic and linguistic resources and models for Romanian were created by SpeeD research group, while the demo application was developed upon Kaldi speech recognition toolkit and Kaldi GStreamer Server.

Rich Speech Transcription Service demo

The Rich Speech Transcription (RST) service is the output of the LVCSR-ROM project carried out by the SpeeD research group between October 2013 and June 2014. The web-service enables individuals to access the textual content of a multimedia document (news bulletin, interview, lecture, meeting recording, etc.) without watching it. This feature is of critical importance in many applications such as multimedia databases indexing and retrieval, real-time radio/TV monitoring, transcription of self-recorded documents, etc.

The Online Speech Recorder

The SpeeD Online Speech Recorder is a bi-product of our LVCSR project which was developed and used to effectively expand the speech database needed to train the Romanian acoustic models.

Voice Login

The Voice Login project aim was to develop an online biometric authentication system with a high degree of security. This system allows users to login into their accounts based on their voice specificities. The Voice Login system performs Speaker Verification to verify that the user is who he claims he is and Automatic Speech Recognition to verify that the user utters specific words and not others. This second feature is needed to prevent malicious users from trying to log in using pre-recorded speech of the claimed user.