VoiceInteraction | VOISCRIPTUM Project

VOISCRIPTUM Project

The main goal of this project is to automate transcriptions of interrogations and depositions for the purpose of maximizing resources and reducing the current burden and time spent in manual transcription procedures from scratch.

Description:
Automatic Speech Transcription to Support Judicial Activity and the Public Ministry. An ongoing project with the participation of the Attorney General's Office, specifically with DCIAP, as a user and receiver of its results. The entire operation is managed internally by the DCIAP team, always guaranteeing all the confidentiality required in this process.

VOISCRIPTUM, a solution based on the latest technologies for audio processing, segmentation and speaker identification, as well as transcription of spoken content, which will serve to support legal proceedings.

The main objective will be the automatic transcription of debriefing sessions and depositions for the purpose of maximizing resources and decreasing the burden and time spent, currently, in manual transcription procedures from scratch. Furthermore, it is intended to equip the entire judiciary with an individually customized dictation system, minimizing repetitive typing tasks.

This project addresses an area in which the procedures inherent in the investigation phase of legal proceedings are, temporarily, very extensive and include several speakers and people of interest, where it will not be possible to follow a personalized approach, leading us to resort to speaker-independent models.

After capturing and recording the audios, they are downloaded to a processing server through a web interface. This is responsible for importing, managing and monitoring the entire automatic transcription process, making the result available for editing purposes.

In this workflow segment, there will be a complementary interface added in order to provide a graphical platform to edit/correct the transcript. This editor has a built-in audio player, enabling a correction process without the need for external tools, as there is a synchronization between what is being heard and what is being corrected.

In a perspective of evolution and constant improvement of the results of automatic transcriptions, the feedback of this “software” will be guaranteed based on the corrections made, where the system itself adapts its acoustic and linguistic models to the domain of Justice.

Additionally, and in the same universe, we will demonstrate, with this innovative project, the capabilities of maximizing people and means offered by automatic speech transcription technology in the particular area of justice.

It is intended that this technology will serve as a basis for the construction of a personal and personalized dictation system, allowing magistrates to create orders and other judicial documents, more quickly and using their own voice. It will be a system adaptable to the user's voice and vocabulary, returning what was said in text form immediately after being spoken. Real data will be collected to create models directly adapted to the field of the judiciary.

Grants: Project code: LISBOA-01-0247-FEDER-027496

Intervention region: Lisbon

Beneficiary entity: Voiceinteraction – Speech Processing Technologies, S.A

Approval Date: 2018-03-13

Start Date: 2017-11-01

Completion Date: 2019-05-02

Total eligible cost: €281.960,42

European Union financial support | FEDER: €112.784,17