Links
The Santa Barbara Corpus of Spoken American English
The Santa Barbara Corpus includes c. 250,000 words of transcribed conversations from all over the U.S. It controls for a number of social variables, such as a speaker’s gender, age, hometown, education, occupation and ethnicity, making it useful for socio-linguistic research. The corpus also forms the main component of the spoken part of the the International Corpus of English U.S. (ICE-US). Sound files are available upon request.
The transcripts include precise time stamps and many indications of intonational properties. An example of a transcription is shown below:
67.880 69.398 .. you probably won't lose any either.
69.398 70.034 [But uh,
69.398 70.441 PAIGE: [@@@ @]
70.034 70.922 KRISTIN: (H)] Okay.
70.922 73.197 So we're gonna take [2eighteen2] hundred ca[3lories3]=,
71.562 72.022 PAIGE: [2(SNIFF)2]
72.624 73.000 [3Uh h-3] --
73.197 73.907 How um,
The Switchboard Corpus
The Switchboard Corpus contains c. 260 hours, more than 2.5 million words, of transcribed telephone conversations. Participants were 543 speakers (302 male, 241 female) from all areas of the United States. A Second phase of the Switchboard Corpus added an additional c. 370 hours, or c. 4 million words, of telephone conversations by 679 speakers predominantly from the Midwest. A part of the corpus, called The Switchboard Dialog Act Corpus, comprising c. 1.4 million words, was made available for free in 2008.
The transcribers followed freely available guidelines. They foresee segmentation of speech into utterances, avoidance of punctuation, "non-speech sounds" such as [throat_clearing], and corrections of mispronunciations like "[splace/space]". An excerpt of a transcript is given below.
sw2930A-ms98-a-0071 269.056375 277.018750 uh-huh i had my boys build their own little c[amp]- you know three leg camp stools by cutting a branch off the tree and binding them together with bark
sw2930A-ms98-a-0072 277.018750 278.342375 [silence]
sw2930A-ms98-a-0073 278.342375 288.195250 and uh we cooked in a pot i baked them a cake right there on a camp fire they were so impressed i used a tire chain and the bottom of a Dutch oven
sw2930A-ms98-a-0074 288.195250 291.936625 [laughter]
sw2930A-ms98-a-0075 291.936625 303.571500 [silence]
sw2930A-ms98-a-0076 303.571500 304.858125 uh-huh
The Buckeye Corpus of Conversational Speech
The Buckeye Corpus contains c. 300,000 words of conversational speech from 40 speakers of American English. All speakers were natives of Columbus, Ohio. The corpus is balanced for gender and age (under 30 and over 40), with ten speakers in each group. The data was collected during a roughly hour-long sociolinguistic interview in 1999-2000. The corpus is easily accessible after registration and the orthographic transcripts for each conversation are based on explicit guidelines (Buckeye Corpus Manual Section 4. Transcribing orthography). An excerpt from a text files looks like this:
yeah
as I was teasing hm
<IVER>
okay
<IVER: how long have you been teaching in columbus?>
twenty eight years
this is my twenty eighth year in Columbus
I have two more years to go
then I get to retire
and then I’ll go retrain
and probably do something else for another twenty years
<IVER>
oh no
The corpus also contains annotations for parts of speech and - especially - phonetic detail! All speech sounds are time-aligned at the word and the phone level. This makes it a key resource for research on reduction, deletion and other features of casual pronunciation. Such a phones file is illustrated below:
0.06 121 dh
0.15 121 ah
0.24 121 k
0.37 121 ae
0.44 121 t;
0.51 121 ih
0.59 121 z
0.70 121 aa
0.77 121 n
0.82 121 dh
0.91 121 ah
0.99 121 m
1.12 121 ae+1
1.19 121 t
