The Kiel Corpus of Spoken German Read and Spontaneous Speech

New Edition, revised and enlarged, 2017

Documentation

Klaus J. Kohler, Benno Peters, Michel Scheffers (Editors)

2

CONTENTS

1 The conceptual frame

1.1 History

1.2 Data recording

1.3 Annotation

2 Directory structure

2.1 Read Speech

2.2 Spontaneous Speech APPOINTMENT-MAKING SCENARIO

2.3 Spontaneous Speech VIDEO TASK SCENARIO 2.4 Duration files

3 File naming conventions

3.1 Read Speech

3.2 Spontaneous Speech

4 File formats

4.1 Signal files 4.2 Text files (orthographic and label) 4.2.1 Orthographic conventions 4.2.1.1 Spontaneous Speech - transliteration files

4.2.2 Label files 4.2.2.1 General principles for segmental and prosodic labels 4.2.2.2 Segmental labels: modified SAMPA 4.2.2.3 Principles of PROLAB and list of prosodic labels

5 Speaker information

5.1 Read Speech

5.2 Spontaneous Speech

6 References

3

1 The conceptual frame

1.1 History The Kiel Corpus of Spoken German comprises two sections – Read Speech and Spontaneous Speech. This New Edition contains the revision of the Kiel Corpus of Read Speech (Kiel PhonDat90/92), previously published on CD-ROM in 1994 (IPDS 1994), and of the Kiel Corpus of Spontaneous Speech (dialogues in the APPOINTMENT-MAKING SCENARIO, Kiel VerbMobil), previously published in three parts on CD-ROM in 1995 – 1997 (IPDS 1995 – 1997). The New Edition also contains additional data in both sections. The Read Speech section consists of various text readings – different sets of sentences and two short story-texts. The text materials are listed in the text file PhonDat_texts.txt in the ‘docs’ directory (see 2.0). Details of material selection and distribution across subjects can be found in Kohler (1992). The VerbMobil dialogues (g-dialogues) were recorded with a technique that opens a channel for only one speaker and blocks the other. It generated clearly delimited dialogue turns for processing in the VerbMobil project, but it limits the study of dialogue interaction. The technique also introduced some artificiality into unscripted communication. For further details see Kohler, Pätzold, Simpson (1995). In the New Edition, the Kiel VerbMobil corpus has been enlarged by including additional g-dialogues not contained in the previous issues, and one further dialogue session f06, recorded within the same APPOINTMENT-MAKING SCENARIO but with channel overlap, prior to the introduction of the separation technique in the g-dialogues. In addition, the VerbMobil corpus has been supplemented by a new corpus recorded with channel overlap in a different VIDEO TASK SCENARIO. Similar but non-identical video material was presented separately to two subjects sitting in different rooms. After the presentation, the subjects discussed differences and similarities of what they had seen and heard. As a basis for data collection, two tapes were spliced together from a number of episodes of the well-known German television series ‘Lindenstraße’. The selection of subjects was guided by two new principles: subjects knew each other and were familiar with the ‘Lindenstraße’ series. The two data sets VerbMobil and VideTask are now sub-sections in the Spontaneous Speech section of the Kiel Corpus of Spoken German. 1.2 Data recording 1.2.1 Read Speech For the recording, the speaker was seated in a sound-treated booth and read the sentences through a window from a monitor that was placed in the adjacent room, where the operator controlled the orthographic presentation of the sentences one by one. The story-texts were presented in two and three parts, respectively, on boards positioned inside the booth. A Neumann U87 cardioid condenser microphone was placed approximately 30 cm from the speaker’s mouth. The microphone signals were amplified by a John Hardy M1 pre-amplifier and recorded on a SONY PCM 2500 DAT-recorder at a sampling rate of 44.1 kHz for PhonDat90 and of 48 kHz for PhonDat92, with 16 bit quantization. They were then digitally transferred to computer hard disk and downsampled to 16 kHz as well as high-pass filtered at 40 Hz. 1.2.2 Spontaneous Speech All recordings were carried out with two dialogue partners seated in different quiet, sound-treated rooms. They were unable to see each other. In the VerbMobil dialogues, they communicated via headsets (Sennheiser HDM 410 or 414), in the VideoTask dialogues via earphones and

4

microphones (Sennheiser MD421-U-4) placed on tables in front of them. This recording environment ensured high quality speech signals, which were digitized at 16 kHz straight to disk using a Loughborough LSI96002 board. In the APPOINTMENT-MAKING SCENARIO of VerbMobil the speaker pairs performed 7 (g-dialogues) or 6 (f-dialogue) appointment tasks. For each g-dialogue task there was a single file which was subsequently de-multiplexed and split into separate turn-files. Splitting into separate turn-files was made possible by recording a constant known signal onto the other channel while a speaker was holding a button pressed. For each VideoTask dialogue and for each of the 6 sub-sessions of the f06 dialogue, signals in each recording channel were digitized at 16 kHz straight to disk yielding a stereo file for each dialogue/sub-session (l*a.wav and f*a.wav, respectively). From this stereo file two mono files were derived, one for each channel ([fl]*a_l.wav, and [fl]*a_r.wav. 1.3 Annotation Annotation of the speech data proceeded through the following series of levels:

(1) Orthographic text, which is given in the Read Speech corpus and is produced for the Spontaneous Speech dialogues by orthographic transliteration, with special symbolizations, enclosed in angled brackets, for phonetic phenomena such as breathing, pauses, clicks, hesitation, laughing, throat clearing; transliteration was done by advanced students of phonetics under supervision by Adrian Simpson, Matthias Pätzold, Benno Peters and Klaus J. Kohler.

(2) Phonematic canonical transcription, automatically generated from the transliteration by the grapheme-to-phoneme-module of the text-to-speech-system RULSYS for German (Kohler 1997).

(3) Segmental labelling, produced by manual time alignment of segmental labels on the basis of the automatically generated phonematic transcription of annotation level (2) (Kohler, Pätzold, Simpson 1995).

(4) Prosodic labelling, using the symbolic system PROLAB, which is based on the Kiel Intonation Model (KIM) (Kohler 1997, 2017).

The core of the New Edition of the Kiel Corpus has been annotated at all 4 levels, and two types of label files are provided: *.s2 with and *.s1 without prosodic labels. In both types of label files aspiration of plosives is segmented separately. In the previous CD-ROM issues of the Kiel Corpus label files with and without aspiration labelling were included, named *.s1h and *.s1. There were then no *.s2 files because prosodic labelling has only been added in the New Edition. Both *.s1 and *.s2 files are now supplied with aspiration labelling, but not marked in the file name extension. Annotation may stop at any level. Some Read Speech data contain only levels (1) and (2) (phonematic transcription) in *.s0 files. Some g-dialogues were only annotated at level (1) and therefore have no label files. In some other g-dialogues annotation stopped at level (3) of *.s1 files. To facilitate identification of these additional data they are stored in separate sub-directories depending on corpus section and annotation levels

• addcorp in Read Speech: with annotation levels (1) and (2) only

• VMaddSeg: 3 additional VerbMobil g-dialogues without annotation level (4)

• VMaddOrt: 8 additional VerbMobil g-dialogues with annotation level (1) only All Spontaneous Speech dialogues have level (1) transliteration files (*.trl), one for each task in the VerbMobil dialogues and one for each complete VideoTask dialogue.

5

Segmental and prosodic labelling was carried out with high precision by phonetically trained students under the supervision of Adrian Simpson, Benno Peters and Klaus J. Kohler. A detailed description of the conventions regarding the transliteration as well as the segmentation of signals and the segmental labelling, applied to the VerbMobil g-dialogues, can be found in Kohler, Pätzold, Simpson (1995). The conventions for segmental labelling are the same in all parts of the Kiel Corpus. They are summarized in 4.2.2.2 below. The conventions for prosodic labelling are outlined in 4.2.2.3. To familiarise labellers with the use of PROLAB a training platform was developed (Peters and Kohler 2004). These training materials are given in the old data structure with the CD-ROM file naming. The table below shows the data volumes of the New Edition of the Kiel Corpus: numbers of male and female speakers and their respective recording times in the various parts of the Read and Spontaneous Speech corpora. Please note that the total number of male/female speakers in the two corpora may differ from the sums of speaker numbers in their sub-corpora because some speakers participated in more than one part of the recordings but are only counted once in the total. Similarly, the total recording times of all speakers in the Spontaneous Speech corpus are adjusted for overlap in the f- and l-dialogues by subtracting the stereo file durations.

nr. of speakers recording times (hrs:min:sec) m f ∑ m f total Read Speech 27 26 53 2:43:52 2:57:30 5:41:22 core (levels 1-4) 27 26 53 1:59:45 2:15:20 4:15:05 extension 14 14 28 0:44:07 0:42:10 1:26:17

addcorp (levels 1-2) 14 14 28 0:44:07 0:42:10 1:26:17 Spontaneous Speech 37 27 64 5:51:16 4:06:49 8:34:35 core (levels 1-4) 22 21 43 3:10:24 3:17:56 5:04:50

VerbMobil 18 13 30 2:16:54 1:54:52 3:56:33 g-dialogues 18 12 30 2:16:54 1:24:26 3:41:20 f06-dialogues (overlap) 0 2 2 0:00:00

0:30:26 0:15:13

(stereo) VideoTask (overlap) 4 8 12 0:53:30 1:23:04

1:08:17 (stereo)

extension 15 7 22 2:40:52 0:48:53 3:29:45 VMaddSeg (levels 1-3) 3 3 6 0:29:21 0:16:47 0:46:08 VMaddOrt (level 1) 12 4 16 2:11:31 0:32:06 2:43:37

6

2 Directory structure

The New Edition of the Kiel Corpus of Spoken German has the following directory structure. 2.0 ‘docs’ directory This directory contains documentation text files that provide important information for the two corpus sections.

Read Speech Info_KielCorp_2017.pdf PhonDat_texts.txt PDspeakers.txt Aipuk26.pdf Aipuk29.pdf

Spontaneous Speech Info_KielCorp_2017.pdf VM/VTspeakers.txt Aipuk29.pdf 2.1 Read Speech +---PhonDat90 | +---berlin | | +---k01 | | | +---k01be001.s[12] | | | +---k01be001.wav . . | | | +---k01be100.s[12] | | | +---k01be100.wav | | +---k02 | | +---k03 | | +---k04 | | +---k05 | | +---k06 | | +---k61 | | +---k62 | | +---k63 | | +---k64 | | +---k65 | | +---k66 | +---marburg | | +---k07 | | | +---k07mr001.s[12] | | | +---k07mr001.wav . . | | | +---k07mr100.s[12] | | | +---k07mr100.wav | | +---k08 | | +---k09

7

| | +---k10 | | +---k11 | | +---k12 | | +---k61 | | +---k62 | | +---k67 | | +---k68 | | +---k69 | | +---k70 | +---butter | | +---k22 | | | +---k22butt1.s[12] | | | +---k22butt1.wav | | | +---k22butt2.s[12] | | | +---k22butt2.wav | | | +---k22butt3.s[12] | | | +---k22butt3.wav | | +---k23 | | +---k24 | | +---k25 | | +---k26 | | +---k27 | | +---k28 | | +---k29 | | +---k30 | | +---k61 | | +---k62 | | +---k76 | | +---k77 | | +---k78 | | +---k79 | | +---k80 | +---nordwind | | +---k13 | | | +---k13nord1.s[12] | | | +---k13nord2.s[12] | | | +---k13nord1.wav | | | +---k13nord2.wav | | +---k14 | | +---k15 | | +---k16 | | +---k17 | | +---k18 | | +---k19 | | +---k20 | | +---k21 | | +---k61 | | +---k62 | | +---k71 | | +---k72 | | +---k73

8

| | +---k74 | | +---k75 | +---restkorp | | +---k61 | | | +---k61cn001.s[12] | | | +---k61cn001.wav . . | | | +---k61cn020.s[12] | | | +---k61cn020.wav | | | +---k61ko001.s[12] | | | +---k61ko001.wav . . | | | +---k61ko063.s[12] | | | +---k61ko063.wav | | | +---k61s1001.s[12] | | | +---k61s1001.wav . . | | | +---k61s1045.s[12] | | | +---k61s1045.wav | | | +---k61s2001.s[12] | | | +---k61s2001.wav . . | | | +---k61s2025.s[12] | | | +---k61s2025.wav | | | +---k61sr001.s[12] | | | +---k61sr001.wav . . | | | +---k61sr015.s[12] | | | +---k61sr015.wav | | | +---k61tk001.s[12] | | | +---k61tk001.wav . . | | | +---k61tk030.s[12] | | | +---k61tk030.wav | | +---k62 | +---addcorp not labelled, *.wav and *.s0 with orthographic text and phonematic transcription | | +---k13 . . | | +---k30 | | +---k71 . . . | | +---k80

9

+---PhonDat92 | +---erlangen | | +---dlm | | | +---dlmer001.s[12] | | | +---dlmer001.wav . . | | | +---dlmer100.s[12] | | | +---dlmer100.wav | | +---hpt | | +---kko | | +---rtd | | +---uga | +---siemens | | +---dlm | | | +---dlmsi001.s[12] | | | +---dlmsi001.wav . . | | | +---dlmsi100.s[12] | | | +---dlmsi100.wav | | +---hpt | | +---kko | | +---rtd | | +---uga Key berlin 100 Berlin sentences: be marburg 100 Marburg sentences: mr butter The Butter Story in 3 parts: butt1/2/3 nordwind The North Wind and the Sun in 2 parts: nord1/2 restkorp 20 CNET sentences: cn 63 Kohler sentences: ko 45 SEL sentences: s1 25 additional SEL sentences: s2 15 Schiefer/Sommer sentences: s3 30 Tillmann/Kohler sentences: tk erlangen 100 Erlangen sentences: er siemens 100 Siemens sentences: si

Each of the terminal nodes in the tree is a directory containing the signal and segmentation files of one speaker who was recorded speaking that particular part of the corpus. Two speakers (k61=kko and k62=rtd) produced the whole corpus. 2.2 Spontaneous Speech APPOINTMENT-MAKING SCENARIO The following dialogues constitute the core of this sub-section of the Spontaneous Speech: g07 – g12, g14, g19, g21, g25, g31, g37, g38, g41, g42 with channel separation, and f06 with channel overlap.

There is a directory for each dialogue with 7 or 6 sub-directories for the 7 or 6 sub-dialogue tasks in the g-dialogues and in the f06 dialogue, respectively. Each sub-directory contains the

10

transliteration file (*.trl) of the sub-dialogue, the signal files and their label files for each dialogue turn/chunk (*axy[lr].wav/s1/s2), and, in the case of the overlapping f-dialogue, the dual and single channel signal files (f*a.wav, f*a_[lr].wav l = left, r = right) as well as the label files for each channel (f*a_[lr].s1, f*a_[lr].s2). This directory structure is displayed in the VerbMobil overview below. xy numbering of turns/chunks starts at 00, instead of 000 in the previous CD-ROM edition; the free space in the name is used for marking the channel as l or r. For the links of channels to speaker names see 5.2. +---VerbMobil | +---g07a | | +---g071a | | |+---g071a.trl | | |+---g071a00r.s1 | | |+---g071a00r.s2 | | |+---g071a00r.wav | | |+---g071a01l.s1 | | |+---g071a01l.s2 | | |+---g071a01l.wav | | +---g072a . . | | +---g077a | +---g08a . . | | +---f061a | | |+---f061a.trl | | |+---f061a.wav | | |+---f061a_l.s1 | | |+---f061a_l.s2 | | |+---f061a_l.wav | | |+---f061a_r.s1 | | |+---f061a_r.s2 | | |+---f061a_r.wav | | |+---f061a00r.s1 | | |+---f061a00r.s2 | | |+---f061a00r.wav | | |+---f061a01l.s1 | | |+---f061a01l.s2 | | |+---f061a01l.wav | | +---f062a . . | | +---f066a The VMaddSeg dialogues g36, g43, g44 have no *.s2 label files in the otherwise identical directory structure. The VMaddOrt dialogues g20, g27, g28, g29, g30, g39, g40, g45 have no label files, only the *.trl and the various *.wav files in the same directory structure.

11

2.3 Spontaneous Speech VIDEO TASK SCENARIO

The directory structure and the file naming follow the same principles as for the overlap f-dialogue. Since there are no sub-dialogues, and in order to create the same directory structure depth as in the g- and f-dialogues, a sub-directory level 1 is inserted, which then contains all the files, including the *.trl file for the complete dialogue.

+-- VideoTask | +---l01a | | +---l011a | | |+---l011a.trl | | |+---l011a.wav | | |+---l011a_l.s1 | | |+---l011a_l.s2 | | |+---l011a_l.wav | | |+---l011a_r.s1 | | |+---l011a_r.s2 | | |+---l011a_r.wav | | |+---l011a00r.s1 | | |+---l011a00r.s2 | | |+---l011a00r.wav | | |+---l011a01l.s1 | | |+---l011a01l.s2 | | |+---l011a01l.wav | +---l02a | | +---l021a . . | +---l06a 2.4 Duration files Directories containing audio recordings now also supply text files with the extension ‘.dur’ in which recording durations are listed. These files are intended to facilitate the conversion of label files to e.g. 'praat TextGrid' format, where this information has to be included in the header. Each line in such a file contains the base-name of the audio file and, separated by tab stop, its duration in seconds. The accuracy of the duration value is such that multiplying it by the sampling rate (16000 Hz), then rounding it to an integral value, will give the exact number of samples in the audio file. The last line contains a 3-field summary: group identification, number of files and summed duration in seconds, again separated by tab stops. For Read Speech the group identification consists of the 3-character speaker identification code (e.g. ‘k12’) followed by the abbreviate form of the sub-corpus, e.g. ‘be’ for ‘berlin’ or ‘rest’ for ‘restkorp’. For Spontaneous Speech, it consists of the sub-dialogue/task code (e.g. ‘g123a’), followed by either ‘_l’ or ‘_r’, identifying the speaker. In both cases, the base-name of the duration file matches this group identification. In Spontaneous Speech files with temporal overlap, the summary line refers to the full-length mono recording files.

12

Example of Read Speech k21add.dur k21be001 2.4000000 k21be002 1.9200000 k21be003 2.9440000 k21be004 2.4000000 k21be005 2.4000000 k21ko029 2.0480000 k21ko030 4.5440000 k21ko031 3.2320000 k21ko032 2.4000000 k21ko033 2.5920000 k21ko034 3.5200000 k21ko035 2.9440000 k21ko036 3.3280000 k21ko037 2.4960000 k21ko038 2.1120000 k21ko039 2.2080000 k21ko040 2.9440000 k21ko041 2.6880000 k21ko042 2.5920000 k21ko043 2.7840000 k21ko044 4.6400000 k21ko045 1.8560000 k21ko046 3.3280000 k21ko047 2.4960000 k21ko048 2.9440000 k21ko049 3.5200000 k21ko050 3.8080000 k21ko051 2.0480000 k21ko052 1.8560000 k21tk002 3.1360000 k21tk003 2.9440000 k21tk004 3.0400000 k21tk005 5.2800000 k21tk006 2.4000000 k21tk007 2.7840000 k21tk008 4.8000000 k21tk009 2.7840000 k21tk010 5.5680000 k21tk011 3.0400000 k21tk012 3.7120000 k21tk013 3.0400000 k21tk014 2.2080000 k21tk015 4.0640000 k21tk016 4.8000000 k21tk017 4.2560000 k21tk018 3.8080000 k21tk019 4.2560000 k21tk020 4.6400000 k21tk021 2.3040000 k21tk022 3.1360000 k21tk023 2.0480000 k21tk024 2.4960000 k21tk025 3.6160000 k21tk026 3.6160000 k21tk027 2.6880000 k21tk028 3.2320000 k21tk029 3.6160000 k21tk030 3.2320000 k21add 58 181.5360000

13

Example of Spontaneous Speech without overlap g123a_r.dur g123a01r 3.6495000 g123a03r 7.2637500 g123a05r 1.3469375 g123a07r 2.1277500 g123a08r 5.4131875 g123a10r 5.5070000 g123a12r 11.9478750 g123a14r 1.5536250 g123a16r 7.6930000 g123a18r 1.5526250 g123a20r 6.3388750 g123a22r 0.7286250 g123a_r 12 55.1227500 Example of Spontaneous Speech with overlap f064a_l.dur f064a01l 29.6234375 f064a03l 27.4740625 f064a04l 22.0065000 f064a_l 3 79.1040000 3 File naming conventions

3.1 Read Speech k05be005.s? speaker (PhonDat90) ^^^ k05be005.s? sentence/text type: ^^ be Berlin sentences mr Marburg sentences butt The Butter Story nord The North Wind and the Sun cn CNET sentences ko Kohler sentences s1 SEL sentences s2 additional SEL sentences sr Schiefer/Sommer sentences tk Tillmann/Kohler sentences dlmsi001.s? speaker (PhonDat92) ^^^ dlmsi001.s? er Erlangen sentences ^^ si Siemens sentences k05be005.s? sentence number or text section ('The Butter Story' is in 3 ^^^ parts, e.g. k24butt2.s1, 'The North Wind and the Sun' in 2 parts, e.g. k74nord1.wav) k05be005.s1 file; labels in SAMPA; plosive release segmental label ^^ (+ aspiration) labelled separately k05be005.s2 and prosodic label file; segmental labels in SAMPA; ^^ segmental plosive release (+ aspiration) labelled separately

14

k05be005.s0 not phonetically labelled, only orthographic text and ^^ phonematic transcription k05be005.wav signal file; 16-bit/16kHz ^^^ 3.2 Spontaneous Speech The names of the transliteration, signal and label files are constructed according to the following conventions: [fgl]071a05r.s1 dialogue scenario ^^^ f, g = APPOINTMENT-MAKING g = button pressing l = VIDEO TASK g0?1a05r.s1 dialogue sitting ^^ g07?a05r.s1 dialogue task number (1-6)in f, (1-7) in g, ^ inserted sub-directory 1 in l g071a05r.s1 recording institute (a = Kiel) ^ g071a05r.s1 turn number in g, chunk number in f06 or l ^^ g071a05[lr].s1 recording channel, l = left, r = right (links to ^^ speaker names in 5.2) g071a05r.s[12] label file s1 = segmental labels, s2 = segmental ^ ^^ and prosodic labels; segmental labels in SAMPA; plosive release (+ aspiration) labelled separately g071a05r.wav signal file of turn/chunk (16-bit/16kHz) ^^^ g071a.trl transliteration file of sub-dialogue (task) in f ^^^ or g, and of complete dialogue in l [fl]0?1a.wav dual channel signal file of sub-dialogue (task) in ^^^ f, and of complete dialogue in l (16-bit/16kHz) [fl]0?1a_[lr].wav single channel signal file of one speaker of ^^ ^^^ sub-dialogue (task) in f, and of complete dialogue in l (16-bit/16kHz), l = left, r = right(links to speaker names in 5.2) [fl]0?1a_[lr].s[12] label file for single channel signal file of one ^^ ^ ^^ speaker of sub-dialogue (task) in f, and of complete dialogue in l, l = left, r = right (links o speaker names in 5.2) segmental labels in SAMPA; plosive release(+ aspiration) labelled separately

15

4 File formats

4.1 Signal files Signal files (*.wav): 16-bit, 16 kHz, RIFF WAVE format 4.2 Text files (orthographic and label) Text files: 7-bit ASCII files, DOS format (CR/LF at line end) 4.2.1 Orthographic conventions The orthographic text (given for Read Speech or transliterated for Spontaneous Speech) follows DUDEN before the spelling reform. In the text files of the Kiel Corpus, word-initial small letters are kept sentence-initially, and TEX conventions are used to represent umlaut and ‘ß’: "a, "A = ä, Ä; "o, "O = ö, Ö; "u, "U = ü, Ü; "s = ß, thus excluding the use of double inverted commas for quotations. However, in the text materials speakers were given to read out these transformations were not made (see PhonDat_texts.txt in the /kiel-corpus/docs directory). 4.2.1.1 Spontaneous Speech - transliteration files All the turns of a g-dialogue task are contained in one transliteration file (extension ‘.trl’) stored in the dialogue task sub-directory. Each turn in the transliteration file has an index comprising three letters (speaker name) and a 3-digit numbering (beginning at 000). The old CD-ROM 3-digit numbering of turns in the transliteration files is thus kept, as against the change to the 2-digit numbering of turn label files. The text of each turn is indented 8 spaces and an empty line separates each turn from the next. Global commentaries are placed after ; at the beginning of a line. See Kohler, Pätzold, Simpson (1995) for further details, especially on the transliteration, between angled brackets, of non-lexical signal categories, such as hesitations, articulatory productions like laughing, throat clearing, hesitational lengthening, breathing and pauses as well as external noises. Example ;Dialog: G071A ;Zuletzt bearbeitet am: 23.05.2017 TIS000: ja , guten Tag . dann fange ich einfach mal an und wollte Sie mal fragen , wie das aussieht/- <"ahm> wir m"u"sten also insgesamt drei<Z> Arbeitssitzungen festlegen . zwei davon m"u"sten wir zweit"agig machen . <"ahm> wann h"atten Sie denn daf"ur mal Zeit ? HAH001: ja <A> , also f"ur den eint"agigen , wenn wir den als <:<Lachen> erstes:> erledigen wollten , quasi , w"are mir ganz recht Montag der achte November . TIS002: <A> Montag , der achte , das sieht bei mir ganz schlecht aus , denn die Woche vom Samstag , dem sechsten , an , da bin ich weg . <"ahm> die Woche davor , wie s"ahe das aus , erste Novemberwoche ? . . For each of the l-dialogues and the f06 sub-dialogues orthographic transliteration was carried out on the dual channel recording. Initially a dialogue was segmented into consecutive dialogue sections of approximately 30 sec, bounded by silence in both channels. The speech sections from both speakers in these chunks were then transliterated and numbered consecutively throughout the

16

(sub-)dialogue, in such a way that the section from the speaker who started speaking first in each chunk was given the next consecutive number first. All the transliterated sections for each sub-dialogue in f and for each complete dialogue in l were entered into one transliteration file, which is stored in the task sub-directory of the f-dialogue or in sub-directory 1 of the l-dialogues. The indexing of sections, the formatting and the transliteration conventions are the same as for the transliteration of the non-overlapping g-dialogues. Example ;Dialog: L011A ;Zuletzt bearbeitet am: 30.11.2000 CJE000: also , +/mein/+ meine Folge hatte gar keinen <:<#> Namen:> . +/das/+ die ging einfach los . <A> die bei dir hie"s sie richtig so ? nee , bei <:<#Kopfh> mir/-:> <P> m . <P> <Ger"ausch> ja , <P> ja . <:<#Kopfh> <A> <Schlucken> <A>:> ja Valle . <P> m , mit diesem Typ . <A> genau , die hat ihm das gesteckt , da"s das Kind nicht von ihm ist , und da hat er sie die Treppe runtergeschubst , ne ? <A> .. TEV001: <Schmatzen> <A> <:<#> ach so:> <A> . <A> ja , bei mir wird 's richtig <"ahm> <A> +/aufge=/+ <P> d=/+ also , gezeigt , ja . <A> ja , und dann <A> , ja , <%> wie fing 's dann an , dann fing 's an hier mit , <A> na , wie hei"st sie noch ? <Lachen> Walze , ne ? <A> Walze +/mit dem/+ <P> mit dem/+ <P> mit dem anderen da im Bett . <A> .. TEV002: ach so , das hab' ich nicht gesehen <A> . <Lachen> <A> aber ej sag' mal , ich denk' , das ist ja Phonetik und nich<Z>t Psychologie . <P> <Ger"ausch> <P> und was ist das denn ? na ja , <A> gut . <P> also/- <A> <Ger"ausch> <Schmatzen> <A> ach , das hab' ich nicht gesehen . <P> .. . . 4.2.2 Label files The segmental transcription is modified SAMPA, the prosodic annotation follows PROLAB: cf. Kohler, Pätzold, Simpson (1995) for segmental labelling; Kohler (1997), Peters and Kohler (2004) for PROLAB labelling. Both segmental and prosodic labelling conventions are summarized in this section. From orthographic representations of the sentences, texts and dialogues, phonematic, citation-form word transcriptions were automatically generated in modified SAMPA, including punctuation marks. Subsequently, segmental and prosodic labelling was carried out in xassp (Scheffers and Rettstadt (1997); Kohler, Pätzold, Simpson (1997)). This procedure produced a segmental label file *.s1 for each sentence or text section from the phonematic transcription by allocating one symbol after another to a point on the time scale of the acoustic signal file, and modifying the series of segmental labels. A segmental label referring to a signal portion is bounded by the time mark of the next label, i.e. segmental labelling is strictly linear. Labels without duration are put on the same time point as the next label that is associated with a time extension. Then time-aligned but durationless prosodic labels were added to the segmental label files *.s1, creating *.s2 files. For the labelling of transliterated non-lexical material and external noises see Kohler, Pätzold, Simpson (1995).

17

4.2.2.1 General principles for segmental and prosodic labels (1) All segmental and prosodic labels (including syntactic punctuation) are placed on one tier.

• Segmental labels in lexical items are prefixed by ## at word boundaries, e.g. ##f in fort, fahren $# at word boundaries inside compounds, e.g. $#f in fortfahren $ in all other cases, e.g. $t in fort; the prefixes are in turn followed by % if the segment boundary is uncertain.

• # prefix of non-verbal labels and syntactic punctuation marks, e.g. #p: (pause), #h: (breathing), #c: (beginning of sentence), #. (period)

• & indicates prefixes of prosodic, and only prosodic, labels, and is put after # in the default case, e.g. #&2^ (accent level 2 with medial peak) $ inside lexical items, e.g. in the case of stress shift, see PEAK Examples in 4.2.2.3 (4).

(2) Label prefixes ## , # and $ are obligatory in the label lines of the *.s[12] files. In the running transcription section of these files, ## and # are replaced by two blanks, $ by one (see 4.2.2.4).

(3) Prosodic labels (as well as punctuation marks) are linked to a point in time, but have no duration. They receive the same time stamp as the subsequent label. So if there are several such labels in a row they are all put on the same time as the next label that is associated with a time extension. There are ordering conventions for prosodic labels, see section 4.2.2.3 B.

(4) Uncertainty as to the prosodic category to be labelled is marked by %, placed after the prosodic index &, e.g. &%PGn (uncertain prosodic phrase boundary). In the case of two prosodic features in one label, each may be marked by %, e.g. &%2^ or &2^% or &%2^%.

4.2.2.2 Segmental labels: modified SAMPA The phonematic canonical transcription is based on a phonemic system of German which recognizes 46 units

24 vowels 20 oral, and 4 nasal vowels, which may occur in loan words from French

22 consonants 3 (x C Q) have non-phonemic status, using them avoids the need to include morphological information to account adequately for the differences in the intervocalic dorsal fricative in pairs such as Frauchen[] and rauchen []or the presence vs. absence of glottalization in pairs such as vereisen []and verreisen []Vowels

SAMPA IPA Example a: a k'a:m k'am kam Kamm e: E b'e:t b'Et Beet Bett i: I r'i:t r'It riet ritt o: O b'o:k b'Ok bog Bock u: U b'u:s@ b'Us@ Buße Busse y: Y h'y:t@ h'Yt@ Hüte Hütte

18

E: k'E:z@ Käse 2: 9 h'2:l@ h'9l@ Höhle Hölle a~ rEsto:r'a~ Restaurant E~ t'E~ Teint O~ se:z'O~ Saison 9~ pa6f'9~ Parfum aI tsv'aI zwei aU b'aUx Bauch OY n'Oyn neun @ l'e:z@n lesen 6 l'e:z6 Leser h'e:6 h'E6 her Herr Consonants

SAMPA IPA Example b b'aIn Bein d d'aIn dein f f'aIn fein g g'ICt Gicht h h'aIn Hain j j'a: ja k k'aIn kein l l'ICt Licht m m'aIn mein n n'aIn nein p p'aIn Pein r r'aIn rein s l'o:s Los t l'o:t Lot v v'aIn Wein z z'aIn sein C z'ICt Sicht x z'Uxt Sucht S S'aIn Schein Z l'o:Z@ Loge N z'INt singt Q Q'aIns eins

19

Label Modifications

In segmental labelling, one of four decisions has to be taken:

(1) the phonematic symbol is aligned without modification

(2) the phonematic symbol has no corresponding signal portion, it is marked as deleted, and is put on the same mark as the next non-deleted label: e.g. $@- (3) the phonematic symbol does not represent the signal portion adequately, it is marked as replaced by a phonetically more appropriate symbol from the canonical list: e.g. $n-m (4) there is no phonematic symbol on offer for the signal portion to be labelled, a label is inserted: e.g. $-t . Biphonematic symbol sequences of a monophthongal vowel and the vowelled realization 6 of post-vocalic /r/ are amalgamated to a single label, which is aligned with the beginning of the entire vocalic portion of the signal, e.g.

Phonematic Label Example E 6 $E6 Herr e: 6 $e:6 Heer 2: 6 $2:6 hört

This is done to avoid problems in attempting to segment such vocalic portions into two parts and to align the two symbols with the beginning of each. It follows from this procedure that in the phonematic canonical section of *.s[12] files the two symbols are separated by a blank, but they are on one label line. However, if the sequence is disyllabic, the label is split up and the second part is marked by = on a separate label line, e.g. ##f $r $'y: $=6 früher Replacements do not introduce symbols in addition to those contained in the canonical lists of vowels and consonants; they index changes to other categories within the vowel or the consonant system. In the case of vowels, only replacements by lax vowels, monophthongization or schwa are marked. Insertions applying to all data

In addition to the insertion of symbols from the canonical list, such as $-t in ##Q $a $l $-t $s , the following segment-related additional labels occur: $-q Creaky voice (a) in addition to, or instead of, a glottal stop: ##Q $-q ##Q- $-q (b) to mark a morpheme boundary or the separation of vowels where a glottal stop is not symbolized in the phonematic transcription: ##n $a $x $#-q $aI $n $#-q $a $n $d $6 nacheinander ##Q- $-q $E $v $E $n $t $-th $u: $-q $'E $l eventuell (c) to replace lenis or fortis plosives in nasal or lateral contexts: ##k $9 $n $t-q $-@ $n+ könnten

(d) to insert glottalization instead of a velar plosive that is not represented canonically ##f $6 $b $'I $n $d-n $U $N $-kq ##n $a: $x+ Verbindung nach

20

$-~ Nasalization in the context of a deleted nasal consonant: placed after the latter if nasalization only affects the post-nasal sequence, or before it if it affects the preceding material: ##Q- $U $-~ $n- $s+ uns ##Q- $E-@ $s+ ##n- $-~ $I C $t-+ es nicht. -MA Presence of phonetic correlates of deleted labels in preceding context: ##v $a:6-a $#S $'aI $n $-MA $l- $I- $C- ##Q- $aI- $n-m ##b $'I $s $C $@- $n wahrscheinlich ein bißchen palatality of deleted syllable -lich transferred to preceding nasal The label is put at the beginning of the deleted sequence to indicate a phonetic residue in the segment(s) preceding it. $-z: hesitational lengthening in speaking, put after the lengthened segment. In all the above cases, the inserted label (together with all subsequent deleted labels in the case of –MA) is put on the same time mark as the next non-deleted segmental label and therefore has no duration. There are, however, also segmental insertions that do have durations:

$-?h plosive release phase: burst and any aspiration ? = [ptkbdg] linked to the preceding stop phase Labelling of spoken deviations from orthographic text in Read Speech n: “non-text”: deviation from orthographic text in reading. Three cases are distinguished:

##-n: marks a stretch of speech that is not represented orthographically and is not segmented and labelled, e.g. false starts

k078butt1: ##-n: ##Q $'aU $s #f $6 $k $-kh $"aU $f $t speaker erroneously reads ##f $6 of verkauft, then corrects herself and reads the actual orthography ausverkauft; the false start is labelled as ##-n: with the duration of the signal chunk

##n: marks a word as deviant in reading, e.g. slip of the tongue, and is put at the beginning of

the misread item, which is segmentally labelled; it has no duration but occurs on the same time mark as the first segment which now gets a segment label prefix $

k03be014 ##n: $n-d $-dh $e: $b-m $@- $%n-m+ ##d-n $e: $m+ ##t $-th $'E $l $6 in neben dem Teller the initial n is replaced by d, probably triggered by the subsequent initial oral t of Teller between two accented syllables; ##n: marks the deviant pronunciation of the following word

##n:- marks an orthographic word that is not pronounced k022butt2: ##n:- $Q- $a- $l- $s-+ ##%S $-MA $o:- $n+ ##-n: instead of the written sequence als schon the speaker produces schon als; the phonematic labels of als are marked as deleted, and the deviant reading is indicated by a preceding durationless ##n:-. The duration of the signal stretch corresponding to the actually pronounced als after schon is represented as ##-n: and the stretch is not segmented (see above).

21

Whereas pauses and breathing are annotated in Spontaneous Speech transliteration files for automatic conversion to canonical transcription files and subsequent manual labelling, this symbolisation is absent from orthographic text in Read Speech, and therefore has to be inserted and marked as such at the labelling stage when the phenomena occur, by the labels

#-p: ‘pause’ #-h: ‘breathing’

which have the durations of the respective stretches in the speech signal. 4.2.2.3 Principles of PROLAB and list of prosodic labels A Labels for prosodic categories (1) PROSODIC PHRASE BOUNDARIES

• Syntagmatic prosodic phrasing is marked by phrase boundaries at the right edges of the prosodic chunks:

&PGn in the default case &PG/ after truncations and false starts &PG; at a technical break (button pressing) in the appointment-making data

• In the main PROLAB system, prosodic phrase boundaries are not subclassified according to strength of separation. n is a place holder for subsequent finer differentiation.

(2) SENTENCE ACCENT

The lexical items within prosodic phrases receive an ACCENT specification on 4 levels, put before the segmental sequence

&2 default accent &3 reinforced accent &1 partial deaccentuation &0 complete deaccentuation e.g. #&2 Max #&0 hat #&0 einen #&3 Brief #&1 geschrieben. “Max has written a letter.” with narrow focus on Brief (answering “what did he write?”), complete deaccentuation of the

function words “hat” and “einen”, strong prominence on “Max”, and weak prominence on “geschrieben”.

(3) LEXICAL STRESS

Within words, one or more syllables are specified for STRESS.

• It is marked at the segmental level and linked to the symbol for the stressed vowel ' primary stress " secondary stress e.g. ##f $'O6 $t $#f $"a: $r $@ $n fortfahren

• Function words, identified by + postfixed to the last segment, by default do not get a lexical stress mark, e.g. ##m $I $t+ mit.

• If function words receive a sentence accent, the double stress mark $' ' is inserted before the accented vowel, e.g. ##m $' ' $I $t+ 'mit.

• The same stress marking applies in the case of stress shift in relation to the canonical transcription, e.g. ##v $ 'U $n $d $6 $b $' ' $a:6 wunder'bar.

22

(4) PITCH PATTERNS

Each ACCENT > &0 receives a distinctive pitch pattern which is either (rising-)falling – PEAK or (falling-)rising – VALLEY or FLAT and is hooked to the environment of the stressed syllable of the accented word, more

particularly to its stressed vowel. It consists of two parts

• the SYNCHRONIZATION of the f0 maximum/minimum with the stressed vowel

• the CONTINUATION of the pattern beyond the synchronization point to concatenate with the next ACCENT pattern or to terminate at the next PHRASE BOUNDARY if there are no fur- ther accents in the prosodic phrase.

SYNCHRONIZATION is marked by an addition to the ACCENT label, CONTINUATION by the

insertion of another prosodic label before the next ACCENT or PHRASE BOUNDARY label. In each case, the labels are specific for each of the three pitch patterns.

• PEAK

SYNCHRONIZATION &2) EARLY synchronization of the peak maximum before the stressed vowel &2^ MEDIAL synchronization of the peak maximum in the stressed vowel &2( LATE synchronization of the peak maximum late in, or after, the stressed vowel, similarly for &1 and &3 CONTINUATION &2. fall to low pitch, creating a large pitch dip between two accents or ending in the

speaker's lowest voice range at the end of the prosodic phrase &1. fall to medium pitch between two accents or at the end of the prosodic phrase &0. no pitch dip between two accents (e.g. ‘hat pattern’ between two peaks) or ending

level at the prosodic phrase boundary &2., or &1., combined contours at phrase boundaries: fall is followed by low rise &2.? or &1.? combined contours at phrase boundaries: fall is followed by high rise &2; or &1; pseudoterminal contours at phrase boundaries: very slight rise after fall,

softens terminality without removing it altogether. Examples #&0 am #&2^ Samstag #&0 oder #&1. #&2( Sonntag #&2. #&PGn “on Saturday or Sunday” either – or alternatives #&0 am #&2^ Samstag #&0 oder #&0. #&2) Sonntag #&2. #&PGn “on Saturday or Sunday = at the weekend” #&2^ tschüß #&1., #&PGn "bye" #&2^ ##t $-th $E6 $m $'i: $n $&0. $&2^ $#v $' ' $"Y $n $S $&1. $&2^ $' ' $@ #&1.

Ter'min'wün'sche “preferences for dates”, pronounced with three accents: the first is hooked to the primary-stress syllable min of the canonical form and given a #& ACCENT label preceding the segmental string; the other two require primary lexical stress positions to be inserted into the canonical form by $' ' , to which $& ACCENT labels can then be linked inside the lexical item.

23

• VALLEY

SYNCHRONIZATION &2] EARLY synchronization of the valley minimum before the stressed vowel &2[ LATE synchronization of the valley minimum in the stressed vowel similarly for &1 and &3 CONTINUATION &, low rising &? high rising, ending in the speaker's high voice range Examples #&0 er #&0 ist #&0 in #&2] Rome #&? #&PGn “he is in Rome?” asking for

confirmation #&0 er #&0 ist #&0 in #&3[ Rome #&? #&PGn “he is in Rome?” surprise question #&0 er #&0 ist #&0 in #&2[ Rome #&, #&PGn “he is in Rome.” friendly non-

categorical statement

• FLAT pattern

SYNCHRONIZATION &2- &1- &3- CONTINUATION &0. &1. &2. Examples #&2^ heute #&0 ist #&1. #&2( schönes #&0. #&2- warmes #&0. #&2) Sommerwetter

#&2. #&PGn “today we have got beautiful warm summer weather” #&0 er #&0 hat #&0 einen #&2( Brief #&2. #&1- geschrieben #&2. #&PGn “he has written a letter”

(5) DOWNSTEP and UPSTEP, RESET

In a sequence of peak contours of accent level 2 within a prosodic phrase, the maximum f0 is by default successively lowered. This is traditionally handled by DECLINATION over time. KIM treats it as structurally controlled DOWNSTEP from accent to accent, irrespective of the time that elapses between them. This default DOWNSTEP is not labelled. On the other hand, if the DOWNSTEP is interrupted within a prosodic phrase an UPSTEP is labelled by putting | in between the prosodic prefix and the ACCENT marker, e.g. &|2^ .

Across PHRASE BOUNDARIES there is default f0 RESET, i.e. DOWNSTEP starts afresh. If this is

not the case and DOWNSTEP continues across the prosodic phrase boundary it is marked as &=PGn .

(6) PREHEAD

Unaccented syllables before the first PEAK ACCENT in a prosodic phrase are the PREHEAD of the contour and may be low or high in relation to the PEAK maximum. A low PREHEAD is taken to be default and not labelled. A HIGH PREHEAD is labelled as one of two different types

&HP1 flat f0 at level of PEAK maximum &HP2 high f0 onset, gradually descending to level of PEAK maximum.

24

(7) RATE and REGISTER

In the labelling of the APPOINTMENT-MAKING SCENARIO, deviations from an impressionistically judged mean speech rate of a speaker are labelled as

&RP speeding up &RM slowing down; deviations from an impressionistically judged mean pitch level of a speaker are labelled as

register changes &LR change to a lower pitch level &HR change to a higher pitch level These changes within a speaker's delivery are in general associated with prosodic boundaries

and therefore marked at the beginning of prosodic phrases in which they take place. These labels are only provided in the g and f dialogues.

B Ordering of prosodic labels Punctuation marks, labels for CONTINUATION, for PHRASE BOUNDARIES, for sentence beginning, for HIGH PREHEAD, RATE, REGISTER, for type of ACCENT are put in this fixed order on the same time mark as the word boundary segment, also at word boundaries inside compound words. In the following listing, alternatives at the same structural point are enclosed between { }. #{, . ? !} punctuation mark #&{0. 1. 2. , ? etc.} CONTINUATION part of PITCH PATTERN #&{PGn =PGn PG/ PG;} PHRASE BOUNDARY; #&PG; for technical breaks in the APPOINTMENT-MAKING

SCENARIO #c: sentence beginning #&{HP1 HP2 RM RP LR HR} HIGH PREHEAD, RATE, REGISTER; the ordering among these

labels is free #&{1 2 3}{( ^) [ ]} ACCENT, SYNCHRONIZATION of PITCH PATTERN ## word boundary $&{0. 1. 2.} CONTINUATION part of PITCH PATTERN inside compound $&{1 2 3}{( ^) [ ]} ACCENT, SYNCHRONIZATION of PITCH PATTERN inside

compound $# boundary of compound element $' ' inserted lexical stress before vowel to receive ACCENT 4.2.2.4 Label file structure Label files have the following structure:

a. file name

b. 7-bit orthographic representation of the text (sentence, story section, turn, chunk) bounded by 'oend' at the beginning of the next line

c. phonematic canonical transcription of the turn bounded by 'kend' (‘kanonisch’) at the beginning of the next line

d. segmental transcriptions (modified SAMPA) and prosodic annotations (PROLAB) of the text constructed from the labels placed by the segmenter, bounded by 'hend' at the beginning of the next line

e. label line with time stamp (sample number), label, starting time in sec

25

Examples of Read Speech file type *.s2 with segmental and prosodic labelling k01be004.s2 "Uber die Felder weht ein Wind . oend Q y: b 6+ d i:+ f 'E l d 6 v 'e: t Q aI n+ v 'I n t . kend c: &0 Q- y: b 6+ &0 d -dh i:+ &2( f 'E l d -dh 6 &0 v 'e: t -th &0 Q- -q aI n+ &0. &2) v 'I n t -th . &2. &PGn hend 3225 #c: 0.2015000 3225 #&0 0.2015000 3225 ##Q- 0.2015000 3225 $y: 0.2015000 3847 $b 0.2403750 4741 $6+ 0.2962500 5837 #&0 0.3647500 5837 ##d 0.3647500 6472 $-dh 0.4044375 6666 $i:+ 0.4165625 7570 #&2( 0.4730625 7570 ##f 0.4730625 9175 $'E 0.5733750 10460 $l 0.6536875 11506 $d 0.7190625 11657 $-dh 0.7285000 11926 $6 0.7453125 13715 #&0 0.8571250 13715 ##v 0.8571250 14870 $'e: 0.9293125 16309 $t 1.0192500 16814 $-th 1.0508125 16958 #&0 1.0598125 16958 ##Q- 1.0598125 16958 $-q 1.0598125 16958 $aI 1.0598125 18948 $n+ 1.1841875 20015 #&0. 1.2508750 20015 #&2) 1.2508750 20015 ##v 1.2508750 21173 $'I 1.3232500 22161 $n 1.3850000 23749 $t 1.4842500 24322 $-th 1.5200625 24730 #. 1.5455625 24730 #&2. 1.5455625 24730 #&PGn 1.5455625 file type *.s1 with segmental labelling k01be004.s1 "Uber die Felder weht ein Wind . oend Q y: b 6+ d i:+ f 'E l d 6 v 'e: t Q aI n+ v 'I n t . kend c: Q- y: b 6+ d -dh i:+ f 'E l d -dh 6 v 'e: t -th Q- -q aI n+ v 'I n t -th . hend 3225 #c: 0.2015000 3225 ##Q- 0.2015000 3225 $y: 0.2015000

26

3847 $b 0.2403750 4741 $6+ 0.2962500 5837 ##d 0.3647500 6472 $-dh 0.4044375 6666 $i:+ 0.4165625 7570 ##f 0.4730625 9175 $'E 0.5733750 10460 $l 0.6536875 11506 $d 0.7190625 11657 $-dh 0.7285000 11926 $6 0.7453125 13715 ##v 0.8571250 14870 $'e: 0.9293125 16309 $t 1.0192500 16814 $-th 1.0508125 16958 ##Q- 1.0598125 16958 $-q 1.0598125 16958 $aI 1.0598125 18948 $n+ 1.1841875 20015 ##v 1.2508750 21173 $'I 1.3232500 22161 $n 1.3850000 23749 $t 1.4842500 24322 $-th 1.5200625 24730 #. 1.5455625 file type *.s0 without segmental but with phonematic labels k14be004.s0 "Uber die Felder weht ein Wind . oend Q y: b 6+ d i:+ f 'E l d 6 v 'e: t Q aI n+ v 'I n t . kend hend 1600000 #c: 99.9999375 1600000 ##Q 99.9999375 1600000 $y: 99.9999375 1600000 $b 99.9999375 1600000 $6+ 99.9999375 1600000 ##d 99.9999375 1600000 $i:+ 99.9999375 1600000 ##f 99.9999375 1600000 $'E 99.9999375 1600000 $l 99.9999375 1600000 $d 99.9999375 1600000 $6 99.9999375 1600000 ##v 99.9999375 1600000 $'e: 99.9999375 1600000 $t 99.9999375 1600000 ##Q 99.9999375 1600000 $aI 99.9999375 1600000 $n+ 99.9999375 1600000 ##v 99.9999375 1600000 $'I 99.9999375 1600000 $n 99.9999375 1600000 $t 99.9999375 1600000 #. 99.9999375

27

Example of Spontaneous Speech VerbMobil g-file g077a08r.s2 TIS008: ja , ich bedank' mich . oend j 'a: , Q I C+ b @ d 'a N k @ m I C+ . kend c: &2) j 'a: , &2. &PGn &0 Q- -q I C+ &2^ b -bh @ d -dh 'a N k-Q @- &0 m I C+ . &2. &PGn hend 1 #c: 0.0000000 1 #&2) 0.0000000 1 ##j 0.0000000 903 $'a: 0.0563750 2159 #, 0.1348750 2159 #&2. 0.1348750 2159 #&PGn 0.1348750 2159 #&0 0.1348750 2159 ##Q- 0.1348750 2159 $-q 0.1348750 2159 $I 0.1348750 3096 $C+ 0.1934375 3661 #&2^ 0.2287500 3661 ##b 0.2287500 4774 $-bh 0.2983125 4959 $@ 0.3098750 5415 $d 0.3383750 6005 $-dh 0.3752500 6177 $'a 0.3860000 7127 $N 0.4453750 7827 $k-Q 0.4891250 8918 $@- 0.5573125 8918 #&0 0.5573125 8918 ##m 0.5573125 9607 $I 0.6003750 10522 $C+ 0.6575625 12846 #. 0.8028125 12846 #&2. 0.8028125 12846 #&PGn 0.8028125 Example of Spontaneous Speech VideoTask section file l011a00r.s1 CJE000: also , +/mein/+ meine Folge hatte gar keinen <:<#> Namen:> . +/das/+ die ging einfach los . <A> die bei dir hie"s sie richtig so ? nee , bei <:<#Kopfh> mir/-:> <P> m . <P> <Ger"ausch> ja , <P> ja . <:<#Kopfh> <A> <Schlucken> <A>:> ja Valle . <P> m , mit diesem Typ . <A> genau , die hat ihm das gesteckt , da"s das Kind nicht von ihm ist , und da hat er sie die Treppe runtergeschubst , ne ? <A> oend Q a l z o:+ , m aI n+ /+ m aI n @+ f 'O l g @ h a t @+ g 'a: 6 k 'aI n @ n :k n 'a: m @ n . d a s+ /+ d i:+ g 'I N Q 'aI n #f "a x l 'o: s . h: d i:+ b aI+ d i: 6+ h 'i: s z i:+ r 'I C t I C z o:+ ? n 'e: , b aI+ :k m i: 6+ /- p: m . p: g: j 'a: , p: j 'a: . :k h: :k w: :k h: j 'a: v 'a l @ . p: m , m I t+ d i: z @ m+ t 'y: p . h: g @ n 'aU , d i:+ h a t+ Q i: m+ d a s+ g @ S t 'E k t , d a s+ d a s+ k 'I n t n I C t+ f O n+ Q i: m+ Q I s t+ , Q U n t+ d a:+ h a t+ Q e: 6+ z i:+ d i:+ t r 'E p @ r 'U n t 6 #g @ S "U p s t , n @+ ? h: kend

28

c: %Q a %l z o:+ , m '' aI n+ /+ m '' aI n @+ f 'O %l g @ h a t -th @+ g -gh 'a:6 k -kh 'aI n @- %n :k %n 'a: m @- %n-m . c: d a s+ /+ d -dh i:+ g -gh 'I N Q- -q 'aI n #f "a x l 'o: s . c: h: d i:+ b aI+ d i:6+ h 'i: s z-s i:+ r 'I C t -th I C z o:+ ? c: n 'e: , b aI+ :k m '' i:6+ /- p: '' m . c: p: g: j 'a: , p: j 'a: . c: :k h: :k w: :k h: j 'a: v 'a l @ . c: p: '' m , %m I t+ %d -dh i: z @- m+ t -th 'y: p . c: h: %g -gh @ n 'aU , d i:+ h '' a t -th+ Q- i: m+ d a s+ g -gh @ S t -th 'E k %t , %d -dh a s+ d- a s+ k -kh 'I n t n I C t-+ f O n+ Q- '' i: m+ Q- I s t-+ , Q- U n t-+ %d-n a:+ h '' a t -th+ Q- e:6+ z i:+ d i:+ t -th r 'E p @ r 'U n t 6 #g @ S '' "U p s t , n '' @+ ? c: h: hend 187 #c: 0.0116250 187 ##%Q 0.0116250 579 $a 0.0361250 887 $%l 0.0553750 1158 $z 0.0723125 1983 $o:+ 0.1238750 2456 #, 0.1534375 2456 ##m 0.1534375 3401 $'' 0.2125000 3401 $aI 0.2125000 5024 $n+ 0.3139375 6544 $/+ 0.4089375 . . . . . . A speaker’s complete label file l0?1_[lr].s[12] has the same structure as the individual section files of that speaker. The orthographic, phonematic canonical, label and label-line parts of the complete label file are organised such that each contains the linear sequence from all of the speaker’s section files. In the orthographic part, only the identification of the first section is kept. This formatting also applies to complete label files for dialogue tasks in the f-dialogue. Example of Spontaneous Speech VerbMobil f-file with overlap f064a_l.s1 MLG001: <A> <Schmatzen> <#Klopfen> . <Schmatzen> auch<Z> nicht gleich zu Beginn , nein <A> ? <Schmatzen> <#Klopfen> . <Schmatzen> ja<Z> . <:<#Klopfen> <A>:> . <hm> nein<Z> <A> . <P> der f"unfzehnte , sechzehnte , das w"are bei mir m"oglich . <Schlucken> <:<#Klopfen> m:> <P> . ja<Z> . <#> <A> . <Ger"ausch> nja<Z> <A> . <"ahm> . <#Klopfen> . ja<Z> , <P> Montag , der zw"olfte <A> . das w"are eine M"oglichkeit . <:<#Klopfen> <A>:> . ja<Z> , wie w"ar' 's mit dem vierten Oktober <A> ? <P> . ja <A> , wunderbar . <#> <A> . mhm <Schmatzen> . <P> . ja <A> . <#Klopfen> am Donnerstag , den zwanzigsten ? <:<#Klopfen> <A>:> . ja <A> . <P> wunderbar . <A> . ja <A> , das w"are ja schon erledigt <A> . <#Rascheln> . oend h: s: :k p: . s: Q aU x+ z: n I C t+ g l 'aI C t s u:+ b @ g 'I n , n 'aI n h: ? s: :k p: . s: j 'a: z: . :k h: . v: m n 'aI n z: h: . p: d e: 6+ f 'Y n f t s e: n t @ , z 'E C t s e: n t @ , d a s+ v E: r @+ b aI+ m i: 6+ m '2: k l I C . w: :k m p: . j 'a: z: . :k h: . g: n j 'a: z: h: . v: Q E: m . :k p: . j 'a: z: , p: m 'o: n t a: k , d e: 6+ t s v '9 l f t @ h: . d a s+ v E: r @+ Q aI n @+ m '2: k l I C k aI t . :k h: . j 'a: z: , v i:+ v E: r @+ Q E s+ m I t+ d e: m+ f 'i: 6 t @ n Q O k t 'o: b 6 h: ? p: . j 'a: h: , v 'U n d 6 b a: 6 . :k h: . m h m s: .

29

p: . j 'a: h: . :k p: Q a m+ d 'O n 6 s t a: k , d e: n+ t s v 'a n t s I C s t @ n ? :k h: . j 'a: h: . p: v 'U n d 6 b a: 6 . h: . j 'a: h: , d a s+ v E: r @+ j a:+ S o: n+ Q E 6 l 'e: d I C t h: . :k p: . kend c: h: s: :k p: . c: s: %Q -q '' aU x+ z: n I C t+ %g -gh l 'aI C t s u:+ b -bh @ g -gh 'I n , %n 'aI n h: ? c: s: :k p: . c: s: j 'a: z: . c: :k h: -s: . c: v: m %n 'aI n z: h: . c: p: %d e:6+ f 'Y n-m f t s e: n t -th @ , z 'E C t s e: n t -th @ , d -dh '' a s+ v E: r @+ b -bh aI+ m i:6+ m-b -bh '2: k -kh l I C . c: w: :k '' m -h: -:k p: . c: j 'a: z: . c: :k h: . c: g: n j 'a: z: h: . c: v: Q- -q E: m . c: :k p: . c: j 'a: z: , p: m 'o: n t -th a: k-x , d -dh e:6+ t s v '9 %l f t -th @ h: . c: %d -dh '' a s+ v E: r @+ -:k Q- -q aI n @+ m '2: k -kh l I C k -kh aI t -th . c: :k h: . c: j 'a: z: , v i:+ v '' E:-E:6 r- @-+ Q- E- s+ m I t-+ %d -dh e: m+ f 'i:6 t @- n Q- O k -kh t -th 'o: b 6 h: ? c: p: . c: j 'a: h: , v 'U n %d-n 6 b -bh '' a:6 . c: :k h: . c: -g: -p: m h '' m s: . c: p: . c: j -z: 'a: h: . c: :k p: Q- a m+ d -dh 'O n 6 s t -th a: k -kh , d -dh e: n+ t- s v 'a n t- s I C s t @- n -h: ? c: -s: :k h: . c: j 'a: h: . c: p: v 'U n d- -dh 6 b -bh '' a:6 -:k -h: . c: h: . c: j 'a: h: , %d -dh a s+ v E:-E:6 r- @+ j a:+ S o: n+ Q- -q E6 l 'e: d -dh I C t -th h: . c: :k p: . hend 1 #c: 0.0000000 1 #h: 0.0000000 15966 #s: 0.9978125 20722 #:k 1.2950625 20722 $p: 1.2950625 192434 #. 12.0270625 192434 #c: 12.0270625 192434 #s: 12.0270625 194825 ##%Q 12.1765000 195287 $-q 12.2053750 195287 $'' 12.2053750 195287 $aU 12.2053750 198176 $x+ 12.3859375 . . . .

30

5 Speaker information

The tables below provide information about the speakers. Machine-readable metadata are also available in 7-bit ASCII CSV format with tab stops as column separators in the files PD/VM/VTspeakers.txt in the ‘docs’ directory of the Kiel Corpus (see 2.0). 5.1 Read Speech The columns in the table below contain the following information: 1. Speaker abbreviation 2. Gender 3. Age 4. German spoken (classification made by speakers themselves): SH – Schleswig-Holstein NI – Lower Saxony (Niedersachsen) ND – North German (Norddeutsch) NW – North Rhine-Westfalia (Nordrhein-Westfalen) HE – Hessen RP – Rhineland-Palatinate (Rheinland-Pfalz) mix – mixed 5. type of files 6. Time of recording (month/year) 7. duration of recording 8. identity with speakers in other recordings of the Kiel Corpus k01 male 25 SH berlin 09/90 04:35 k02 female 25 SH berlin 09/90 04:29 k03 male 28 SH berlin 09/90 04:04 k04 female 26 SH berlin 09/90 05:40 k05 male 28 SH berlin 09/90 05:42 k06 female 23 RP berlin 09/90 04:41 k07 male 23 SH marburg 09/90 05:04 k08 female 24 SH marburg 09/90 05:30 k09 male 27 SH marburg 09/90 04:55 k10 female 24 SH marburg 10/90 05:08 k11 male 24 SH marburg 09/90 05:20 k12 female 22 SH marburg 09/90 04:11 k13 male 25 SH nordwind 09/90 00:46 k14 female 26 SH nordwind 09/90 00:39 k15 male 25 SH nordwind 09/90 00:49 k16 female 23 SH nordwind 09/90 00:43 k17 male 27 SH nordwind 09/90 00:45 k18 female 28 NW nordwind 09/90 00:43 k19 male 27 SH nordwind 09/90 00:54 k20 female 24 NW nordwind 09/90 00:47 k21 male 26 SH nordwind 09/90 00:40 k22 female 26 SH butter 09/90 01:17 k23 male 24 SH butter 09/90 01:32 k24 female 20 SH butter 09/90 01:17

31

k25 male 27 SH butter 10/90 01:31 k26 female 23 NI butter 09/90 01:17 k27 male 30 SH butter 10/90 01:23 k28 female 25 NI butter 10/90 01:15 k29 male 28 SH butter 09/90 01:42 k30 female 24 NI butter 10/90 01:23 k61 male 55 mix ALL 09/90 26:07 = kko k62 female 33 NI ALL 09/90 23:45 = rtd k63 male 33 SH berlin 09/90 04:30 k64 female 50 Nd berlin 09/90 05:04 k65 male 35 mix berlin 09/90 04:34 k66 female 39 mix berlin 09/90 04:31 k67 male 35 RP marburg 09/90 05:11 k68 female 50 NI marburg 09/90 05:02 k69 male 49 HE marburg 09/90 04:07 k70 female 46 SH marburg 09/90 05:13 = MLG in g41a and

f06a (5.2) k71 male 38 Nd nordwind 09/90 00:44 k72 female 46 NI nordwind 10/90 00:52 k73 male 47 SH nordwind 09/90 00:48 k74 female 56 SH nordwind 10/90 01:03 k75 male 52 Nd nordwind 09/90 00:43 k76 female 30 SH butter 09/90 01:36 k77 male 48 Nd butter 09/90 01:34 k78 female 47 SH butter 09/90 01:18 k79 male 42 SH butter 09/90 01:43 k80 female 48 SH butter 09/90 01:45

PhonDat92 core

dlm male 21 SH erlangen/siemens 0492 13:13 hpt male 24 SH erlangen/siemens 0492 15:01 kko male 57 mix erlangen/siemens 0492 17:23 rtd female 35 NI erlangen/siemens 0492 17:07 uga female 23 SH erlangen/siemens 0492 13:29

PhonDat90 addcorp

k13 male 25 SH addcorp 09/90 03:23 k14 female 26 SH addcorp 09/90 03:09 k15 male 25 SH addcorp 09/90 03:14 k16 female 23 SH addcorp 09/90 03:08 k17 male 27 SH addcorp 09/90 02:57 k18 female 28 NW addcorp 09/90 03:00 k19 male 27 SH addcorp 09/90 04:17 k20 female 24 NW addcorp 09/90 03:35 k21 male 26 SH addcorp 09/90 03:02 k22 female 26 SH addcorp 09/90 02:15 k23 male 24 SH addcorp 09/90 02:29 k24 female 20 SH addcorp 09/90 02:48

32

k25 male 27 SH addcorp 10/90 02:51 k26 female 23 NI addcorp 09/90 02:24 k27 male 30 SH addcorp 10/90 02:31 k28 female 25 NI addcorp 10/90 02:11 k29 male 28 SH addcorp 09/90 02:53 k30 female 24 NI addcorp 10/90 02:38 k71 male 38 Nd addcorp 09/90 02:51 k72 female 46 NI addcorp 10/90 03:09 k73 male 47 SH addcorp 09/90 03:29 k74 female 56 SH addcorp 10/90 05:27 k75 male 52 Nd addcorp 09/90 03:54 k76 female 30 SH addcorp 09/90 02:34 k77 male 48 Nd addcorp 09/90 02:39 k78 female 47 SH addcorp 09/90 02:08 k79 male 42 SH addcorp 09/90 03:37 k80 female 48 SH addcorp 09/90 03:44 = JUM in g36a and in

f06a (5.2) 5.2 Spontaneous Speech The columns in the table below contain the following information:

1. Speaker abbreviation 2. Gender (f= female; m = male) 3. Age 4 German spoken (classification made by speakers themselves) SH – Schleswig-Holstein HH – Hamburg NI – Lower Saxony (Niedersachsen) Nd – North German (Norddeutsch) BE – Berlin NW – North Rhine-Westfalia (Nordrhein-Westfalen) HE – Hessen RP – Rheinland-Pfalz BB – Brandenburg

ST – Saxony-Anhalt (Sachsen-Anhalt) mix – mixed – no information given 5. Profession 6. Dialogue 7. Time of recording (month/year) 8. recording channel: l(eft), r(ight) 9. Duration of recording (stereo overlap or mono channels) 10. identity with speakers in other recordings of the Kiel Corpus

In some dialogues, the speakers were acquainted. This was the case in the f- and l-dialogues, and in g09, g12, g14, g31, g36, g37, g38, g 41, g42.

33

APPOINTMENT-MAKING SCENARIO

Overlap

VerbMobil core f

MLG female 49 SH secretary f06 09/93 l 15:08 = MLG in g41a

= k70 in 5.1

JUM female 51 SH secretary f06 09/93 r 15:08 =J UM in g36a

= k80 in 5.1

Channel separation

VerbMobil core g

HAH male 24 SH student g07 10/93 l 08:12 TIS male 25 NI student g07 10/93 r 09:22

THS male 26 SH student g08 10/93 l 04:57 KAK male 25 SH student g08 10/93 r 07:54

ANS female 24 NI student g09 10/93 l 07:33 FRS female 28 NI student g09 10/93 r 04:07

UTB female 25 SH student g10 10/93 l 08:51 NAR female 25 NW student g10 10/93 r 06:21

JAK male 21 NI student g11 10/93 l 09:48 REK male 25 SH student g11 10/93 r 06:21

OLV male 34 SH student g12 10/93 l 10:23 KAP female 27 NI student g12 10/93 r 06:37

BAC male 24 SH student g14 10/93 l 04:05 SOK male 23 SH student g14 10/93 r 02:44

KAE female 21 Nd physiotherapist g19 10/93 l 06:18 SAR female 19 Nd student g19 10/93 r 04:43

CHD male 24 Nd student g21 10/93 l 06:59 ANL male 25 SH student g21 10/93 r 10:07

WEM female 20 BE student g25 11/93 l 07:40 SIK female 20 Nd student g25 11/93 r 06:22

AME male 19 SH student g31 11/93 l 02:35 SVA male 19 SH student g31 11/93 r 03:14

MEH male 60 mix scientist g37 09/94 l 07:48 ARK male 55 mix professor g37 09/94 r 06:41

FRA female 45 - secretary g38 09/94 l 03:10 HEL male 53 HE professor g38 09/94 r 06:49

34

MLG female 50 SH secretary g41 09/94 l 09:51 = MLG in f06a

= k70 in 5.1

HEE female 57 SH adm.employee g41 09/94 r 12:53

GEP male 58 NI professor g42 10/94 l 08:59 MAW male 60 SH professor g42 10/94 r 19:56

VMaddSeg

JUM female 52 SH secretary g36 09/94 l 08:29 = JUM in f06a

= k80 in 5.1

URG female 56 HH secretary g36 09/94 r 05:38

CHR female 50 Nd employee g43 10/94 l 02:40 ERS male 60 BE professor g43 10/94 r 09:30

THC male 37 SH banker g44 11/94 l 11:09 MAT male 31 SH banker g44 11/94 r 08:42

VMaddOrt

BLA male 26 SH student g20 10/93 l 09:09 PRB male 26 SH student g20 10/93 r 07:48

SIH female 24 Nd student g27 11/93 l 06:33 SOV female 23 SH student g27 11/93 r 06:46

BEK male 36 RP ass.professor g28 11/93 l 04:22 MAK male 25 Nd student g28 11/93 r 04:36

THP male 28 HH scientist g29 11/93 l 10:03 MAL male 34 HH scientist g29 11/93 r 13:19

MAB female 21 HH student g30 11/93 l 06:20 ANM female 20 HH student g30 11/93 r 12:27

HOP male 53 NW scientist g39 09/94 l 11:47 REG male 40 Nd Akad.Rat g39 09/94 r 13:41

HAZ male 49 ST teacher g40 09/94 l 09:19 WAR male 53 BB Wiss.Rat g40 09/94 r 19:26 KAT male 35 SH banker g45 11/94 l 16:49 HOS male 45 SH banker g45 11/94 r 11:12

35

VIDEO TASK SCENARIO

TEV female 23 SH student l01 01/00 l 09:33 CJE female 28 SH student l01 01/00 r 09:33

AHA female 22 SH HS-student l02 02/00 l 13:00 TLU female 20 SH HS-student l02 02/00 r 13:00

APE female 27 SH student l03 01/00 l 09:41 KPA female 32 SH designer l03 01/00 r 09:41

KTH female 23 SH student l04 01/00 l 09:18 CHE female 23 SH student l04 01/00 r 09:18

UHO male 33 SH technician l05 01/00 l 15:22 JKO male 31 SH tradesman l05 01/00 r 15:22

MPI male 35 SH ship-builder l06 10/00 l 11:23 TRA male 31 SH technician l06 10/00 r 11:23

36

6 References IPDS (1994). The Kiel Corpus of Read Speech, Volume 1, CD-ROM#1, Kiel: Institut für Phonetik und digitale Sprachverarbeitung, Christian-Albrechts-Universität.

IPDS (1995). The Kiel Corpus of Spontaneous Speech, Volume 1, CDROM#2, Kiel: Institut für Phonetik und digitale Sprachverarbeitung, Christian-Albrechts-Universität.

IPDS (1996). The Kiel Corpus of Spontaneous Speech, Volume 2, CDROM#3, Kiel: Institut für Phonetik und digitale Sprachverarbeitung, Christian-Albrechts-Universität.

IPDS (1997). The Kiel Corpus of Spontaneous Speech, Volume 3, CDROM#4, Kiel: Institut für Phonetik und digitale Sprachverarbeitung, Christian-Albrechts-Universität..

Kohler, K. J. Webpage. Sound Patterns of German Spontaneous Speech. URL: http://www.ipds.uni-kiel.de/kjk/forschung/lautmuster.en.html

Kohler, K. J. (ed.) (1992). Phonetisch-akustische Datenbasis des Hochdeutschen. Kieler Arbeiten zu den PHONDAT-Projekten 1989 – 1992. Arbeitsberichte des Instituts für Phonetik und digitale Sprachverarbeitung der Christian-Albrechts-Universität Kiel (AIPUK), vol. 26.

URL: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus/docs/Aipuk26.pdf

Kohler, K. J. (1997). Modelling prosody in spontaneous speech. In Y. Sagisaka, N. Campbell, and N. Higuchi (Eds.), Computing Prosody: Computational Models for Processing Spontaneous Speech, pp.187–210. New York: Springer.

Kohler, K. J. (2017). Communicative functions and linguistic forms in speech interaction. Cambridge: Cambridge University Press.

Kohler, K.J., G. Lex, M. Pätzold, M. Scheffers, A. P. Simpson, W. Thon (1994). Handbuch zur Datenaufnahme und Transliteration in TP14 von Verbmobil - 3.0. Kiel: Institut für Phonetik und digitale Spachverarbeitung, Christian-Albrechts-Universität.

URL: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus/docs/Verbmobil-Handbuch_TP14-3.0.pdf

Kohler, K. J., Pätzold, M., and Simpson, A. (1995). From scenario to segment: the controlled

elicitation, transliteration, segmentation and labelling of spontaneous speech. Arbeitsberichte des Instituts für Phonetik und digitale Sprachverarbeitung der Christian-Albrechts-Universität Kiel (AIPUK), vol. 29. URL: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus/docs/Aipuk29.pdf

Kohler, K. J., Pätzold, M., and Simpson, A. (1997). From the acoustic data collection to a labelled speech data bank of spoken Standard German. Arbeitsberichte des Instituts für Phonetik und digitale Sprachverarbeitung der Christian-Albrechts-Universität Kiel (AIPUK), vol. 32, 1-29. URL: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus/docs/kkmpas.pdf

Kohler, K. J., Peters, B., Scheffers, M. (2017a). The Kiel Corpus of Spoken German – Read and Spontaneous Speech. New Edition, revised and enlarged. Data. Kiel: Christian-Albrechts-Universität. URL: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus/

37

Kohler, K. J., Peters, B., Scheffers, M. (2017b). The Kiel Corpus of Spoken German – Read and Spontaneous Speech. New Edition, revised and enlarged. Documentation. Kiel: Christian-Albrechts-Universität. URL: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus/docs/Info_KielCorp_2017.pdf

Peters, B. (2006). Form und Funktion prosodischer Grenzen im Gespräch. PhD Diss., University of Kiel, 2006. URL: http://macau.uni-kiel.de/receive/dissertation_diss_00002078 Peters, B., and Kohler, K. J. (2004). Trainingsmaterialien zur prosodischen Etikettierung mit dem

Kieler Intonationsmodell KIM ("Training Material for Prosodic Labelling with the Kiel Intonation Model KIM"). Kiel: Institut für Phonetik und digitale Spachverarbeitung, Christian-Albrechts-Universität. URL: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus/docs/TrainerA4.pdf URL: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus/docs/KIM.zip

Scheffers, M. and Rettstadt, T. (2000). xassp User's Manual (Advanced Speech Signal Processor

under the X Window System). Kiel: Institut für Phonetik und digitale Sprachverarbeitung, Christian-Albrechts-Universität.

URL: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus/docs/xassp_manual.pdf

38

Queries and comments Any queries or comments are to be directed to: Dr. Benno Peters Institut für Skandinavistik, Frisistik und Allgemeine Sprachwissenschaft (ISFAS) Christian-Albrechts-Universität zu Kiel D - 24098 Kiel Germany email: peters@ipds.uni-kiel.de www: http://www.isfas.uni-kiel.de/de/linguistik/forschung/kiel-corpus Limits of Liability Great care has been taken in the preparation of The New Edition of the Kiel Corpus of Spoken German. The Editors are however not responsible for any damage arising from the use of the Corpus. Copyright notice The New Edition of the Kiel Corpus of Spoken German is copyrighted. It is for research purposes only. No form of commercial use is allowed, nor may the data be sold or transferred, as a whole or in parts, to another party. The files, or any part thereof, may not be reproduced in any form without the written permission from The Editors of The New Edition of the Kiel Corpus of Spoken German Christian-Albrechts-Universität zu Kiel D - 24098 Kiel Germany © Kohler Peters Scheffers 2017