A.I MICRO SERVICES

OXO LYRA: DIALOGUE DETECTOR MICRO SERVICE BASED IN A.I 

The OxoPackage Player Editor  counts with a micro service called Oxo Lyra, developed using machine learning. OxoLyra detects dialogues and creates a perfectly-timed subtitle template that complies with Netflix’s standards. These timed lines are empty, allowing linguists to focus only on transcription and translation tasks, even when working with large-scale projects. With this A.I approach and using GPU, Oxo Lyra takes only 40 seconds to analyze a 90 minute-long feature and to produce a DFXP file with accurate subtitle timing.

Oxobox platforms extract the audio track from the proxy file and move the audio file to the Lyra environment where an inference algorithm analyzes it in and runs a process that looks for the sound of the human voice. The machine learning system of Lyra was trained to “hear” human voice using thousands of segments belong to a hundred hours of series and film features hat was subtitled and QCed by skilled people and create lines of the subtitle empty of text.

Then, these lines are adjusted by a posts process under a set of restrictions and rules present in the Netflix Specs Guide for Timed Text. This post-process is parametric in order to produce different results, depending on the kind of content( I.E: adult or kids content determines the reading speed and, musical may need a different length)

After 20 seconds Lyra delivers an SRT file for a 45 minutes episode that contains a very reliable subtitle that has the right quantity of lines located in the best audio position. Each line has the right length and complains all the Netflix specs regarding duration and separation. 

In this way, it is possible to start to work with human operators with linguistic skills but technical.

These operator has only to fill lines with text and eventually delete lines due to some false positives that the algorithm may produce.