Transcription
The LIUM Speech Project has developed some tools and resources in order to implement an entire automatic speech recognition (ASR) system. The LIUM ASR system is based on the CMU Sphinx 3.3 (fast) decoder. It has participated in the ESTER french evaluation campaign.
While our system is tuned to process with French language, our tools can be used in other languages.
We provide under open source licenses two parts of our system:
- a tool to rescore Sphinx lattices using 4-gram language models (BSD-like license)
- an entire segmentation tool (GPL license)
LIUM distributes some very useful resources which can be used with CMU Sphinx 3.x decoders. They concern only French language, and are distributed under BSD-like license. These resources are:
- a lexicon containing about 65K words
- narrow-band and large-band acoustic models
- trigram and quadrigram language models
References
- P. Deléglise, Y. Estève, S. Meignier, and T. Merlin, The LIUM speech transcription system: a CMU Sphinx III-based system for french broadcast news, in Proc. of Interspeech'05, Lisbon (Portugal), September 2005
- Poster for the above paper