CMU Sphinx The CMUSphinx project is a leading automatic speech recognition project in the open source world. Since being released as open source code in 1999, it provides a platform for building speech recognition applications. It's used in desktop control software, telephony platforms, intelligent houses, computer-assisted language learning tools, information retrieval and mobile applications. Traditionally, CMUSphinx provides support for low-resource and underdeveloped languages.
Inter-speaker variability is an important issue in Automatic Speech Recognition (ASR). ASR systems trained (or adapted) to recognize a single speaker (speaker-dependent ASR) have significantly better performances than ASR systems which aim at recognizing any voice. Usually, speaker-independent ASR systems are backed-up with an unsupervised, online speaker adaptation (MLLR) module to overcome the performance degrading caused by the inter-speaker variability. MLLR is a cheap adaptation method that is suitable when the amount ofdata is limited. It's a good idea to use MLLR for online adaptation. At the moment pocketsphinx and Sphinx4 can use MLLR transforms, but cannot create MLLR transforms! Practically, MLLR transforms can only be created during training, not at runtime. This project aims at implementing a Java module to be used in CMU Sphinx4 for on-line (runtime) adaptation of acoustic models.
First steps: Start User Documentation