Common use of Comparison between native and non Clause in Contracts

Comparison between native and non. native speech Figure 1 presents the percentage of labels whose boundaries are shifted by less than a threshold and compares them for native (FF) and non-native speakers (GF). As expected, we see that the segmentation of non-native speech is slightly more difficult than that of native speech. This is mainly due to the lack of fluency of most non-native speakers, which generates hesitations, insertions of speech segments such as glottal stops and fricatives /?, h/. The presence of erroneous realizations such as aspirated voiceless stop consonants (French voiceless stops are not aspirated) also explains this difference. With CoALT, users can define classes for phones and their context in order to provide shift histograms. We used this feature to investigate the inter-annotator agreement in specific cases.