In view of The Language Technology Industry Summit 2017, we have spoken with Béatrice Bacconnet, CEO at Bertin IT about new opportunities that the latest developments in speech technology bring to the media.
Phlippe Wacker, Secretary General, LT-Innovate: Bertin IT, via its Vecsys brand, is a key player in the speech technology industry, providing speech2text software and services. Could you briefly describe the origins of your company and the main phases of its development up to now?
Béatrice Bacconnet: Bertin IT is the IT pure-player subsidiary of CNIM, an industrial group founded in 1856 and listed on Euronext Paris.
In 2011, Bertin acquired the French renowned specialist in automatic speech processing Vecsys, in the context of a long term collaboration in Defence & Security.
In 2015, keeping on developing into both the public and private sectors, Bertin IT acquired AMI Software, a leading software publisher in market and web intelligence, which it officially launched as a CNIM subsidiary specialized in cybersecurity, digital intelligence and speech solutions, in order to accelerate its growth and break into new markets at an international level.
Today, Vecsys Speech products are packaged within Bertin IT Defence & Security intelligence solutions, but they are also offering more and more dedicated solutions for the Media sector and the Banking & Finance sector, as Speech R&D and product teams are highly-specialized innovation-crazy engineers hitting very impressive new targets !
Phlippe Wacker: Publishing and the media have become major consumers of speech technology and language technology in general. What is driving growth in this sector today?
Béatrice Bacconnet: Several trends are impacting this sector in a short time frame:
Video is gaining a prominent place in media and online communications, so that traditionally written or audio media have to embrace videos and leverage their contents.
At the same time, radio broadcasters are living a revolution, with replay, on-line contents, filmed radio, TV extensions, etc… Some radio groups are making a quantitative leap in terms of media content processing, leveraging and converging all their audio, video and text assets.
A third fortunate trend is that technology and solutions have made enormous progress for this sector ! Bertin IT's speech-to-text solution, MediaSpeech, integrates advanced machine learning and rich enabling services like automatic news learning, elastic cloud, speaker biometric ID, etc.
Phlippe Wacker: Which aspects or challenges of the publishing/media sector offer the best targets for language technology?
Béatrice Bacconnet: For content owners/broadcasters, a main objective is to add value to produced or existing media contents: exposing more info, driving more views, connecting relevant contents together. This is very true for audio and video contents that are usually poorly indexed, but also true for text contents, in terms of categorization, linking, etc.
For media monitoring, a main objective is to automate content scanning, filtering and alert generation as much as possible, while keeping both coverage and relevance at a very high level. Here again, the main challenge is for audio and video contents, but semantic processing like topic detection, relevance scoring, automatic summary, etc. are also looked for.
Phlippe Wacker: What is the impact of AI (Artificial Intelligence) in this industry?
Béatrice Bacconnet: Artificial intelligence (AI) and, more precisely, deep neural networks are setting a fast-paced technological progress in the speech industry so that every release of MediaSpeech pushes back limits in terms of contents, accuracy and processing speed, up to levels we would never have dreamt to achieve a few years ago.
The generalization of machine learning also allows new ways to automate continuous improvement, and embrace many data sources allowing the system to keep on learning.
Phlippe Wacker: What is likely to be the most impactful change in this segment in the next 5 years?
Béatrice Bacconnet: For one, radio mutation and on-line video explosion, but this is just the beginning. Tthen we also expect to see more convergence between broadcast media, on-line media and social media.
Finally, if you combine these with current hardware and software acceleration, the opportunities for real-time processing are huge.
