Vorträge

von Fabian Brackhane und Renate Raffelsiefen
  • Wolfgang von Kempelen's Sprechende Maschine (1791)Fabian Brackhane: Vorträge und Demonstrationen "History of voice synthesis". Amsterdam: The Art of Voice Synthesis, 11. Mai 2016

    Der Versuch, gesprochene Sprache unabhängig von ihrer Erzeugung durch den Menschen und stattdessen durch artifizielle Hilfsmittel zu imitieren oder gar zu synthetisieren, reicht überraschend weit zurück. Spätestens aus dem 18. Jahrhundert sind erste ernstzunehmende Ansätze belegt. Die hierbei unternommenen Versuche greifen natürlicher Weise auf rein mechanische Konstruktionen zurück, die teilweise zu auch heute noch überraschend guten Syntheseergebnissen führen. Bis zur praktischen Nutzbarkeit der Elektrizität zu diesem Zweck um 1930 wurden auch später zahlreiche Versuche der mechanischen Modellierung unternommen. Die Entwicklungen des 18. Jahrhunderts gipfelten in der "Sprachmaschine" des Wolfgang von Kempelen (1734-1804), die erstmals eine grundsätzlich funktionsfähige Rekonstruktion des gesamten menschlichen Sprechapparats darstellt. Zeitgleich veröffentlichte anatomieferne Konzepte (Kratzenstein, Mical) stellten Sackgassenentwicklungen dar, während Kempelens Konstruktion während des 19. Jahrhunderts immer wieder aufgegriffen wurde und auch heute noch ein ergiebiges Forschungsgebiet darstellt.

    Im Rahmen der Vorträge wird zunächst die grundsätzliche Entwicklung der Sprachsynthese von ihren o. g. Anfängen bis in die Gegenwart dargestellt. Daran anschließend wird die - ebenso erstaunliche wie essentielle - Verknüpfung der Sprachsyntheseforschung mit dem Orgelbau behandelt. Die detaillierte Darstellung und Diskussion der Sprachsynthese Wolfgang von Kempelens wird Gegenstand des dritten Vortrages sein.

  • Fabian Brackhane / Renate Raffelsiefen: "Converging evidence for biphonemic diphthongs in German". Manchester: 24th Phonology Meeting, 26.-28. Mai 2016

    Previous phonetic studies of the German diphthongs AI and EU, which contrast in words such as Eile 'hurry' vs. Eule 'owl', agree on recognizing two sequential target positions, connected by both raising and fronting. They further agree on the representation of the respective first targets ([a] for AI versus [ɔ] for EU), but not the second targets (e.g {[ɔoe],[ɔø],[ɔʏ],[ɔy],[ɔɪ], [ɔi],[ɔj]} for EU). Here we argue that the agreement is captured and the disagreement is resolved by recognizing the more abstract targets shown in (1). Significantly, these are precisely the segmental and prosodic representations motivated by strictly phonological evidence (see below). (The third diphthong, AU as in Aule ‚snot’, is given in (1c).)

    (1) a. AI       b. EU       c. AU 

    The representations in (1) account for the properties mentioned above in that the consensus in identifying the first vowel is linked to the nucleus position, a position associated with maximal target approximation, whereas the disagreement is linked to the coda, a position associated with target undershoot (Krakow 1999). Drawing on these correlations between syllabic position and mode of phonetic implementation we conducted four phonetic studies to verify the role of /i/ as a target for the diphthongal offglides. We aimed to determine whether the phonetic realization of coda /i/ indicates a movement towards the phonetic target position observed for /i/ in nucleus position. To base the relevant projections for the offglides on solid ground we measured the formants F1, F2, and F3 in ten equidistant steps from 5% to 95% diphthong duration. Target positions for monophthongs like /i/ were measured in stressed syllabic position in words like sieben. The studies are briefly described below (I – IV).

    I. Comparison of F1/F2-trajectories of AI and EU before various consonants (e.g. before /ç/ Zeichen - Seuchen), based on 50 male speakers of Northwestern Germany. Result: for AI-EU pair the projected F1/F2-trajectories consistently converge near the location of /i/. (Before -/f/ the projected trajectories meet in a point closer to /y/, due to regressive labialization.)

    II. Comparison of F1/F2-trajectories of EU in the word Zeug in male and female speech, based on additional recordings of 50 female speakers of Northwestern Germany. Result: gender-specific differences between the trajectories observed in previous studies (Pätzold & Simpson 1997) follow directly from the different target positions associated with the monophtongs /ɔ/ and /i/ for females and males.

    III. Comparison of F1/F2-trajectories of EU in the word Leute in read versus spontaneous speech based on 320 male speakers from all German speaking areas. Result: significantly closer proximity between the point at 95% diphthong duration for read (versus spontaneous) speech with the target position associated with syllabic /i/.

    IV. Comparison of F3-F2 difference, known to indicate roundedness, for EU in the word Leute in read versus spontaneous speech based on 320 male speakers. Result: significantly stronger unrounding, in accordance with the unrounded target /i/, in read than in spontaneous speech. The last two results strongly support the presence of an actual target which is approximated the more closely, the more care and time is invested in the articulation.

    The phonological motivation for the representations of the offglides with /i/ organized in the syllable margin concerns evidence (much of it novel) pertaining to markedness (e.g. the epenthesis of /i/ as a hiatus breaker in Old High German), neutralization (e.g. the absence of the palatal glide before AI and EU, but not AU, to satisfy a constraint against identical phonemes flanking the nucleus), and correspondence patterns (e.g. the prevalence of „impure rhymes“ involving AI-EU in poetry, due to the same coda vowels in (1a,b), as opposed to the  absence of AI-AU or AU-EU rhymes). It is the convergence of all types of evidence which supports a compelling argument for recognizing the biphonemic representations in (1).