Dynamic pronunciation modeling using phonetic features and symbolic speaker adaptation for automatic speech recognition