RSS News: A Non-acoustic Approach to Crosslingual Speech Recognition Performance Pre

[reza1217]

In this study we introduce an algorithm that objectively predicts the performance of crosslingual acoustic models in speech recognition tasks. This approach does not require the conducting of actual speech recognition tests with target-language speech data for prediction verification; nor does it depend on acoustic measurement techniques. Rather, the algorithm is based on a series of non-acoustic, linguistic metrics characterizing the articulatory phonetic and phonological information of phonemes from both the target and source languages. Because this method can be used to predict the performance of crosslingual models intended for a given target language when target-language speech data is not available for verification, this algorithm is useful both for validating crosslingual model sets for speech recognition applications, and also for making database acquisition decisions that could prove very cost-beneficial.

More...