Phoneless words
There is an interesting phenomenon that has come up in the
Switchboard Transcription Project
where a word that has been plainly audible in the whole utterance
disappears as the time window is restricted, and indeed shows no
visible acoustic correlates in the spectrogram. Presumably, the
combination of linguistic context and timing hints are enough to
`make' us hear the word.
This page illustrates the phenomenon by pointing to some of the examples
from the transcription corpus where we've seen this. The examples were
reported to Eric Fosler by the transcribers. See
Eric for more information...
Examples (from the STP datasets)
- 2523-A-0026 and you_see_now she('s) started a korean health food store
- 2959-B-0001 are you(a) vietnam veteran dudley
- 3353-A-0010 you_know we('d) all go camping my dad(and) my mom and and the kids
compare:
dad(and) my mom
- 3385-A-0001 now then what is your idea on what to serve for(a) dinner party
- 4372-A-0041 accounting department will{pay}(at) the end of the month
- 4718-A-0039 (i)think it_was called credit union news
- 3313-B-0056 try(to) tell my kids
- 3377-B-0045 something(is) coming through now
- 2085-A-0078 more(of) that
- 2309-B-0079 nice talking(to) you also
- 2382-A-0075 the spring creek one is a lot more modern [breath] has (a)lot more
- 4023-B-0050 and kind of a couple(of) people
- 2253-A-0019 i'm embarrassed(to) say the last
- 2776-B-0012 like to cycle (a)lot
- 2982-A-0083 but uh the rest (of)the time i've been an [laughter]
- 3274-A-0024 somebody i talked to last week was_? said they had the uh sil they had problems doing some(of) the work down there because they breath
- 4074-B-0078 so we gave him to um (i)don't know if you've
- 2085-B-0129 you know i make more(than) my boss
- 2874-A-0029 so he was in(a) bad way [static] but uh [static]
- 3801-B-0016 (and)there was no one there but i wanted to be sure
- 3906-A-0008 [static] that's what my [pause] i don't know those are some of my feelings i think free enterprise does it for(a) better price
- 4008-B-0035 find out how reliable the test is you try(to) correlate it with other tests and things of that sort so_it's
- 4147-B-0039 uh [pause] we live right [pause] might say we live right in uh [pause] in the middle of (a)residential area
- 4617-A-0054 maybe they could go with a [pause] with a nine hundred number(and) have people dial in and give their vote
- 2187-A-0094 or [pause] or(a) regular beater
- 2336-B-0114 since [breath] (you)know my parents had_to force me to get my driver's license when i was young
- 2426-B-0048 decided to [pause] carve up that part(of) the world and call part of it [pause] persia
- 3309-B-0021 are into themselves more than they are into (you)know doing something worthwhile for the people they represent
DAn Ellis <dpwe@icsi.berkeley.edu>
International Computer Science Institute, Berkeley CA