This morning, as I wrote about Nemesysco's claim to "detect 'Brain activity finger prints' using the voice as a 'medium' to the brain", I was reminded of something I read last night about a much older forensic application of voice analysis.
In chapter II of Dashiell Hammett's Red Harvest, the Continental Op is questioning the secretary of Donald Willsson, a newspaper editor who's been murdered. The secretary is described as "a small girl of nineteen or twenty". She describes what her boss did during his last afternoon at work.
"You called up -- if it was you he told to come to his house -- at about two o'clock. After that Mr. Donald dictated some letters, one to a paper mill, one to Senator Keefer about some changes in post office regulations, and -- Oh yes! He went out for about twenty minutes, a little before three. And before he went he wrote out a check."
They locate his check book, find the check stub, and establish that Willsson could easily have gotten to the bank and back in 20 minutes.
"Didn't anything else happen before he wrote out the check? Think. Any messages? Letters? Phone calls?
"Let's see." She shut her eyes again. "He was dictating some mail and -- Oh, how stupid of me! He did have a phone call. He said: 'Yes, I can be there at ten, but I shall have to hurry away.' Then again he said: 'Very well, at ten.' That was all he said except, 'Yes, yes," several times."
"Talking to a man or a woman?"
"I didn't know."
"Think. There'd be a difference in his voice."
She thought and said:
"Then it was a woman."
It turns out that she was right, of course. Willsson's phone partner was the memorable Dinah Brand.
Red Harvest was published in 1929. As far as I know, in the intervening 75 years, no one has ever thought to check experimentally what you can tell from one side of a phone conversation about the person on the other side. In particular, the implication of the passage I've quoted is that a listener should (sometimes?) be able to tell the sex of the other person from "a difference in [the] voice". This is a nice Socratic displacement of Holmesian deduction -- the secretary knows how to read the clues, even by reference to a day-old memory, but she has to be reminded of her knowledge.
I'm skeptical that such inferences are generally reliable, in fact. But this is something (else) you could do, when you're forced to listen to someone talking on their cell phone in a public place -- ask yourself whether you have any intuitions about the sex of the person on the other end of the line, just from the "voice" of the side you hear. If you're a bold person, you could check your perceptions by asking for the truth of the matter.
It would be easy enough to test various versions of Hammett's hypothesis in a more conventional sort of experiment.
Overall, it's striking how little we know about most non-lexical aspects of speech perception.
Posted by Mark Liberman at August 29, 2004 10:29 PM