"Yeah day go, baby"

« previous post | next post »

Yesterday, while I was sitting in an interesting session at Speech Prosody 2018, I got a phone call that I didn't answer. The caller left a message that Google Voice transcribed this way:

Lowell is an installer sensor Grace call me. I'll pick it up. That was a break was thinking. Because you had to go to work this morning around, you know, my exact maybe go back to take the brake light. As you said you didn't feel quite right still cyber, even though I was still wearing the back. I might have something. Bye. What thank God. This f****** f*** m*********** train my f****** bank account. What I see your ex. What's your phone number? Yeah day go, baby. Does it have that switch that maybe that's what size over at light source? I'm open. Another f*****. I know what that's like I recognize. Yeah, I was.

Here's the recording:

I gather that this was a wrong number rather than a spam robocall.

What I find interesting about it is that in the stretches of the recording that are unintelligible, the system isn't able to recognize that it doesn't can't actually hear or understand what was said, and instead just hallucinates freely. See "AI hallucinations", 3/10/2018.



15 Comments

  1. outeast said,

    June 14, 2018 @ 7:54 am

    Butt dial? Sounds like two voices having a conversation to me, at some remove from the phone. Although that would presumably mean one has your number and so is at least an acquaintance.

  2. BZ said,

    June 14, 2018 @ 12:18 pm

    The transcription made me think it would be something sexual, but the audio seems to disprove that theory.

  3. outeast said,

    June 14, 2018 @ 1:18 pm

    The transcription made me think it would be something sexual, but the audio seems to disprove that theory.

    It's someone ranting about their wife, isn't it? Drunkenly and unpleasantly, to a rather less drunk buddy. Maybe I'm totally misinterpreting it but I feel more creeped out every time I listen.

  4. ktschwarz said,

    June 14, 2018 @ 1:19 pm

    The comment from "sewmonde" is spam (fake content copied from the post itself, the only purpose is to get the spammer's URL on the page), which is perhaps appropriate for a post about machine-generated language!

    This doesn't look to me like hallucinating freely: it ignores a lot of the input and transcribes only fragments, but the output seems close in sound to those fragments. For example:

    (0:10) Lowell is an installer: I hear "I don't know, man, stall for time"
    (0:22) sensor: I hear "tester"
    (0:25) Grace: I hear "brakes"

    and so on. I'm surprised the transcription didn't put in any symbols like "…" or "___" to indicate sounds that it failed to recognize; that's what my voicemail does.

  5. Michael Watts said,

    June 14, 2018 @ 3:50 pm

    It's also interesting that there's plenty of speech in there that's perfectly intelligible, but which there isn't even an attempt to transcribe. (e.g. around 2:19 "that lady who was behind me this morning (briefly unintelligible) I recognized the plate on the front.")

  6. Noel Hunt said,

    June 14, 2018 @ 5:18 pm

    Mark V. Shaney would appear to be alive and well and now able to vocalize his meanderings.

  7. Sal said,

    June 14, 2018 @ 6:17 pm

    I also thought it sounded like a butt dial overhearing a conversation between two people,. but the topic sounded to me like it was about vehicle repair (at least I thought I heard both "brakes" and "truck" and then the reference to draining his bank account I interpreted as the expense of fixing something).

  8. Nick Barnes said,

    June 15, 2018 @ 3:12 am

    I'm with Sal. It's a butt dial, two guys chatting, partly about vehicle repair. One is much closer to the phone than the other.

  9. Ralph Hickok said,

    June 15, 2018 @ 2:21 pm

    I'm often bemused by the closed captioning (which seems to be on by default) on the video game summaries at MLB.com.

    Here a few samples from the condensed game video of Boston's 2-1 victory over Seattle last night:

    into the corner one nothing Boston dug leak in Iran
    bi devil one-nothing red sucks big fox has that trick trick job Gaby
    we're gonna miss lucky mouth
    that case the price's get fired for the ballgame for tunic red devil plays a girl takes charge throws dug up by Hailey twinkling red off time to to to bugga its deep toward the pen

  10. D.O. said,

    June 15, 2018 @ 3:07 pm

    Are the stars courtesy of Google Voice? That's the only reasonable guess. Prof. Liberman would not censor explicit language on the blog and people do not speak in stars yet, unless Prof. Liberman got a call from a cartoon character…

    It should be reasonably easy for Google (and other similar companies) to estimate the level of certainty of their captioning (or translation, for that matter) and allow the user (customer?) to select a comfort level for the service from "give me your wildest guess" to "ignore if not certain". Should I patent this option and sell it for a vacation house on Virgin Islands?

  11. Michael said,

    June 15, 2018 @ 4:00 pm

    The very end of the call is amusing. It sounds to me like speaker #1 notices that his phone is connected, asks who's on the line, then disconnects when there's no answer.

  12. Maneki Neko said,

    June 15, 2018 @ 7:48 pm

    Definitely car trouble. At 1:23 they discuss a problem with a switch found in older cars.

    A: You know that switch you hit with your foot, remember them?
    B: Yeah, step on it, the high beams come on.
    A: Yeah, yeah, yeah.
    B: Tap it again…
    A: Yeah, they go off.
    B: Yeah, the high beams go off but the low beams don’t work at all

  13. Ray said,

    June 16, 2018 @ 6:30 am

    "yeah day go, baby" = "yeah, there you go, baby"

  14. fev said,

    June 16, 2018 @ 12:14 pm

    Billy is a handyman. Porcelain is the surest plan.

  15. ajay said,

    June 18, 2018 @ 7:47 am

    Agree with Maneki Neko. Also the clearest parts are "because you had to go to work this morning", something about "sensor" and the "drain your bank account" part. "Maybe that switch is bad".
    It clears up at the end as they switch topic – "that lady was behind me this morning, I recognise the plate on the front" and then something about "you've got three quarters, he's got the last one, he's got it, he's good to go".
    Strikes me it might be two men driving in a vehicle? They're talking about the vehicle, then they comment on the traffic they see, then they start rooting around for change for a toll.

    It's someone ranting about their wife, isn't it? Drunkenly and unpleasantly

    Seems like it has a Rorschach quality to it…

RSS feed for comments on this post