Speech Trax: A Bottom to the Top Approach for Speaker Tracking and Indexing in an Archiving Context

Speech Trax: A Bottom to the Top Approach for Speaker Tracking and Indexing in an Archiving Context F´elicien Vallet1, Jim Uro2, J´er´emy Andriamakaoly3, Hakim Nabi1, Mathieu Derval1, Jean Carrive1 1Institut National de l’Audiovisuel, 4 avenue de l’Europe, 94366 Bry-sur-Marne Cedex, France 2Universit´e de Technologie de Compi`egne, rue du Dr Schweitzer, 60200 Compi`egne, France 3T´el´ecom ParisTech, 46 rue Barrault, F-75634 Paris Cedex 13, France firstname.lastname@{ina.fr1, etu.utc.fr2, telecom-paristech.fr3 } Abstract With the increasing amount of audiovisual and digital data deriving from televisual and radiophonic sources, professional archives such as INA, France’s national audiovisual institute, acknowledge a growing need for efficient indexing tools. In this paper, we describe the Speech Trax system that aims at analyzing the audio content of TV and radio documents. In particular, we focus on the speaker tracking task that is very valuable for indexing purposes. First, we detail the overall architecture of the system and show the results obtained on a large-scale experiment, the largest to our knowledge for this type of content (about 1,300 speakers). Then, we present the Speech Trax demonstrator that gathers the results of various automatic speech processing techniques on top of our speaker tracking system (speaker di- arization, speech transcription, etc.). Finally, we provide insight on the obtained performances and suggest hints for future improvements. Keywords: Demonstration, Audiovisual Indexing, Speech Processing, Speaker Tracking, i-Vectors, TV and Radio shows 1. Introduction INA1, France’s national audiovisual institute is the world’s leading source of digitized audiovisual content. It collects and preserves 80 years of radio archives and 70 years of television programs that form the French collective mem- ory. The missions of archives such as INA being to store but also describe and enhance content, the need for auto- matic structuring methods is of utmost importance. If stor- age is not the main issue anymore, indexing methods re- main ill-suited for the retrieval of information in multime- dia databases. Thus, an important human effort for descrip- tion is required to enhance the stored data. In this context, many research works aim at automatically organizing and structuring large quantities of audiovisual documents. Speech processing technologies allow the ex- traction of valuable information from the audio signal. Dis- crimination of speech and music zones, automatic segmen- tation in speaker turns or automatic speech transcription, there are nowadays many tools available. Among these, speaker tracking – the task of finding spoken segments of a particular speaker for which some training material is given – is of great interest. Indeed, famous voices haunt the au- diovisual heritage. Politicians, athletes, intellectuals, an- chormen, etc. our broadcast history resonates with heroic, tender, comical or dramatic accents and the ability to re- trieve such persons’ interventions in the archive is of great value. 2. Related works As stated in (Kinnunen and Li, 2010) speaker tracking is derived from the larger field of the speaker recognition technologies that has been a very hot research topic for several decades. However, the challenge of dealing with “big data” is just getting tackled. For instance, in (Jeon …

Content truncated to 500 words