Comparison of Prosodic Features in Swedish and Finnish Ids/ads Speech

A set of speech corpora, spoken in both IDS and ADS modes containing similar sentences in terms of complexity, was collected in Sweden and Finland in 2007. Prompts for each language were designed by selecting 10 object words and 10 carrier sentences for each language to produce approximately 100 different sentences. Each sentence was spoken 10 times in adult-directed speech (ADS) mode and 10 times in infant-directed speech (IDS) mode, for a total of 2000 sentences per speaker. Four speakers, typically two pairs of married couples from each language, were used to produce a total of 8000 utterances per language. The corpora, that also included Dutch and English, were collected to aid in the exploration of new architectures for speech recognition, as is currently taking place in the ACORNS project This study describes the main prosodic differences found in IDS/ADS speech from the collected Swedish and Finnish material. Statistical analysis of the following main features are reported: F0 histograms: these may reveal speaking style differences in the F0-contours. Spectral tilting and its fluctuation: this also is affected by the voice source and the “softer” speaking style encountered with IDS. Segmental durations and their variations: average speaking rate and indications of monotonic/non-monotonic speaking style. Differences in spectral properties: the signals are analyzed in frames of 10 ms and the MFCC-coefficient vectors clustered in a vector space. The cluster centres reveal average spectral properties and differences in speaking style that can be detected as differences in the number of clusters found as well as differences of their mean vectors. The analysis is performed for four Swedish and four Finnish speakers (two males and two females for each language). Tests are carried out to detect any common characteristics between IDS and ADS speech for these two languages. 1 ACORNS (Acquisition of Communication and Recognition Skills), an EU-funded Sixth Framework Programme, Priority [2], Future and Emerging Technologies, project, 2007-2009.