Connected Digits Recognition Task: ISTC CNR Comparison of Open Source Tools

EVALITA is a recent initiative devoted to the evaluation of Natural Language and Speech Processing tools for Italian. In this work, the results of three open source ASR toolkits will be described. CSLU Speech Tools, CSLR SONIC, CMU SPHINX are applied on the EVALITA clean and noisy digits recognition task and this report will describe the complete evaluation methodology. CSLR SONIC has resulted to have the best performances in all the tasks and even with high specialized trainings. We think that it is mostly because of the PMVDR features used in this system. CMU SPHINX has been the easiest system to train and test and its general performances are only slightly lower than SONIC. CSLU Speech Tools is the most specialized recognition system on digit and its score stands in the middle of the others. Overall, the three systems have Word Accuracy score over 90%.