About the Innovator
The Language and Speech Technology (LST) team at University of Le Mans was set up with the main objective of developing methodologies and technologies for identifying and aggregating data presented as unstructured information in sources of very different nature (video, image, audio, speech, text and social context).
The LST team is a part of LIUM, the Laboratory of Computer Science of the University of Le Mans, a French public institution. The LST team is composed of about 30 members, including 12 permanent staff and they focus their research activities on speech recognition, speaker identification, machine translation, and spoken language understanding.
What is the innovation
The LST team succeeded in developing a software technology for speech and speaker recognition that reduces by a factor of 25 the computation time required to process huge amount of audio and video content, while maintaining the quality of the output. With such a decrease in computation time, this innovation reduces by around a factor 25 the costs of such speech processing. This state-of-the-art speech processing system was built around the LIUM_SpkDiarization and the Kaldi open source toolkits, and has been completely modified thanks to the performances provided by deep neural networks. This work was undertaken in the FP7 EUMSSI project.
Out of the lab – Into the Market
This innovation has been quickly brought out of the lab through the spin-off company Voxolab that benefits from technology transfer agreements with the University of Le Mans. By integrating this innovation into its speech processing architecture, Voxolab has drastically reduced its production cost and can offer to its customers less expensive services.
Benefits of participation in the Framework Programme
LST's participation in FP7 through the EUMSSI project has permitted it to consolidate its visibility at international level on the domain of speech processing. By integrating the EUMSSI consortium, the LST team have developed its international professional network, combined to funding that helped it to consolidate its research activities. Thank to this, the team has successfully participated in different international evaluation campaigns in speech processing, and published a high number of scientific publications in international journals and conferences, in addition to technology transfers.