A corpus of read and conversational Austrian German

Rethinking classification results based on read speech, or: why improvements do not always transfer to other speaking styles

“…Given that in conversational speech the number of phones realized canonically is much lower than in read speech (c.f. Schuppler et al 2014), the number of incorrect labels in training and test data is higher. For this same reason, the creation of broad phonetic transcriptions is a much more difficult task for conversational speech than for read speech, which is for instance also reflected by substantially higher inter-labeler disagreement [5.6% for read speech vs. 21.2% for conversational speech (Kipp et al 1996(Kipp et al , 1997].…”

“…Even though the main reason for using AFs instead of phones is that AFs have more potential for capturing pronunciation variation, most investigations on AF classification have been carried out on read speech, while conversational speech is (far) more prone to pronunciation variation than read speech (Schuppler et al 2014). Even though it is well known that TIMIT (Garofolo 1988), a read speech corpus of American English, is non-generic, it continues to be the most popular corpus for research into AF classification American English (e.g., Pernkopf et al 2009;Pruthi and Espy-Wilson 2004;Siniscalchi et al 2007).…”

An analysis of prosodic boundary detection in German and Austrian German read speech

“…We tested the system on read speech from two corpora, the Kiel Corpus of Spoken German [23], which contains speech from, mainly, Northern Germany, and the Graz Corpus of Read and Spontaneous Speech (GRASS; [3,24]), which contains speech from eastern-Austrian speakers. As these corpora were annotated with different methods (manual phonetic segmentations for the Kiel corpus vs. semi-automatic segmentations for GRASS) and wanting to have comparable input data for a more accurate comparison, we created automatic segmentations for both corpora, using MAUS [1].…”

Prosodic Effects on Plosive Duration in German and Austrian German

“…This study investigates the behavior of German plosives and their neighboring segments at phrase boundaries. In doing so, we analyze data from two varieties of German (Northern German and Austrian German) of different standard pronunciations of the fortis and lenis plosives [10] and different phonological coarticulation processes [11].…”

“…and Spontaneous Speech (GRASS) [21]. We chose these corpora because the speakers are typical of two different varieties of German (i.e., (northern) German and Austrian German), and because there is a substantial overlap in the read material collected in the corpora, making a relatively direct comparison possible.…”