未登录,请登录后再发表信息
最新评论 (0)
播放视频

发掘你的声音

Unlocking the secrets hidden inside your voice

很多人都知道他们的推特和照片墙被监控着
Most people know that their tweets and Instagram posts are monitored.
我们自愿地向世界倾泄大量的个人信息
We voluntarily dump a ton ofpersonal data to the world,
这些信息对广告商 科学家 执法机构来说
and that’s valuable foradvertisers, scientists,
具有极大价值
law enforcement, you name it.
但只有少数人会考虑到个人数据的另一种形式
But few people think aboutanother form of personal data
这种形式会泄露更多信息
that can be even more revealing.
它无处不在 识别度很高 并且难以伪装
It’s ubiquitous, highlypersonal, hard to fake,
它还可能对我们不利
and could be used against us.
你现在所听到的便是它
You’re listening to it rightnow,
你可能很想吐槽我有气泡音吧
and you’re probably going to comment saying that I have vocal fry.
这正是我们的嗓音
It’s our voices.
嗓音会暴露很多我们的:我们的性格
Voices reveal a lot aboutus: our personality,
健康状态 我们的心情
our health, our mood.
而由于亚马逊回声智能音响
And thanks to smart assistants,
谷歌智能家居等智能助手的出现
like the Amazon Echo or Google Home,
我们渐渐习惯了整天对这些设备说话
we’re getting used to talkingto devices all the time.
盖洛普最近的一次投票显示
A recent poll from Gallup found that
约22%的美国人
about 22 percent of Americans
使用智能助手
use smart assistants.
这些公司自然是希望我们
Naturally, companies want us to
能更多地与这些设备交流
talk to the devices even more,
且不知不觉中交出我们声音中的秘密
and potentially give up the secrets in our voices.
因此获悉谈话内容的关键实际上不在于我们说了什么
So the key to the insights we’re talking about isn’t actually what we say,
而在于我们是怎么说的
but how we say it.
也就是说 不是讲话的内容
Meaning not the content of speech,
也不是讲话的用词 而是物理上人说话时
not words of speech, the physics of speech
喉部产生的波形
created from the waveformsfrom the human voice box.
查尔斯 麻玛是纽约大学医学院的
Charles Marmar is the chair of the psychiatry department
精神病学系主任
at New York University Medical School.
他已经对创伤后应激障碍进行了数十年研究
He’s been studying posttraumatic stress disorder
并且已经做了许多工作
for decades and has done a lot of work
来对PTSD的声音生物标记进行定义
in defining vocal biomarkers for PTSD.
这些是从你的声音中提取信息
These are the buildingblocks for pulling meaning
所需的基本要素
out of your voice.
从最广泛层面上来说 生物标记包括
At the very broadest level,biomarkers can involve things
诸如我们声音的音色 语速
like the tone of our voice,the speed we talk at,
强调的用法 以及话语间的停顿
the way we emphasize words,and the pauses between words.
但这仅仅只是开始
But that’s just the beginning.
查尔斯通过运用机器
Using machine learning,
对PTSD患者讲话进行了研究
Charles has studied the speech of people with PTSD
又发现大约18处细微标志
and found about 18 much more nuanced markers
我们甚至可能都没有注意到这些
that we might not even notice.
它们是微小的生物物理学特征
They’re tiny biophysical features,
但是它们一般代表
but in general, they represent speech
那些语调很平 非情绪化
which is flatter, less emotional,
且变化更少的讲话
and less variable.
这一技术不仅适用于PTSD
This wouldn’t only work with PTSD, either.
还能够运用于很多种其他情况
The same technique could beused with many conditions,
且每种都能有一套专属的声音生物标志
and each could have its ownset of unique vocal biomarkers.
我认为像抑郁症和焦虑的其他表现形式
I think for depression, for other forms of anxiety,
可能是躁郁症 尤其是狂躁症
probably for bipolardisorder, particularly mania,
可能会使你大声喧哗 富有攻击性 高音量
where you would have loud,highly forceful, high-volume,
和喋喋不休的表现
high-intensity speech,
这时声音中某些特征的改变意味着
there might be some characteristic changes in voice that signal
此人有患精神疾病的风险
that someone’s drifting towards a psychotic risk.
目前查尔斯的研究依赖于高质量的录音
Right now, Charles relies onvery high-quality recordings
和强大的计算机系统
and a powerful computer system,
但是希望有一天它能改进
but the hope is that one day it can be streamlined
变得更简单 大部分人都能使用
into something simple thatlots of people can use
并且从中受益
and find helpful.
一种以手机应用或云端形式呈现
Something simpler that could be used in the form
且更易于操作的东西
of an app or somethingthat would be in the cloud.
这不仅仅是个假想
That’s not just a theoretical idea.
事实上 一家名为CompanionMx的公司已经开发了
In fact, one company calledCompanionMx has already made
一款心情分析应用 基本用法如下
a mood analysis app. Basically,
病人将声音日志录入应用
patients recordaudio diaries into the app,
人工智能会分析录音并共享结果
an AI analyzes them and shares the results
给病人和他们的医生
with the patient and their doctor.
该应用目前的作用在于 当病人病情好转
And what we’ve seen thisdoes is now the patient,
他们能够知道自己正在好转
if they’re doing better, theyknow they’re getting better,
并且亲眼看到自己在好转
they can see that they’re getting better,
同样的 当情况不怎么乐观的时候 他们也能看到
and they can also see when things are not going so well.
虽然目前该应用只有临床医生能够使用
The app is only available to clinicians right now,
但我们请到CompanionMx公司总裁Sub Datta
but we asked CompanionMx chief executive, Sub Datta,
来给我们讲解该应用是如何运行的
to tell us how it works.
病人需要留下一份声音日志
The patient is required to leave an audio diary,
基本上就是对着应用
which is basically wherethey speak to the app
或手机说一段话
or speak to the phone
至少每周一次
at least a minimum of once a week
每次不少于10秒钟
and for a minimum of 10 seconds.
这大概需要7天时间
And it takes about seven days
来获取基础数据
to capture the baseline data and after that,
然后开始分析数据的特征
analysis of the display of data starts.
因此我们能看到这个软件的作用
So what we’ve seen as a result of that
就是将病人与临床医生拉得更近
is this brings the patient and theclinician together more,
同时病人也会更加投入
and the patient becomes significantly more engaged.
但是除了医学应用 一些商家也正在申请
But beyond medicine, somebusinesses are applying
将该技术应用到一些 令人担忧的方面
the technology in ways thatstart to feel… concerning.
由一名以色列人创办的Voicesense公司承诺
An Israeli startup calledVoicesense promises
能对从某位顾客的投资风格 到员工的雇佣
to do voice analysis on everything from a customer’s investment style
以及营业额等任何方面进行声音分析
to employee hiring and turnover.
其中有一例是 它对债务人的语音录音
In one example, it ran its algorithm
进行了运算分析
on recordings from debtors.
结果十分有趣
The results were intriguing.
它预测的低风险人群中只有 6%违约了
Six percent of the predicted low-risk people defaulted
与之相对的高风险被试者中却有27%
compared to 27 percent of the high-risk subjects.
另一项实验则关注于
Another experimentlooked at the probability
员工可能辞职的几率
that employees would leave.
低风险员工中 有大约13%的人辞职了
About 13 percent of the low-risk workers left
与之相对的高风险组却是39%
compared to 39 percent of the high-risk group.
Voicesense公司声称
Voicesense claims it can
可以通过声音创建一份全面个人特征档案
create a full personality profile using voice,
这个应用程序比查尔斯的PTSD筛查项目更有野心
an application that is far more ambitious than Charles’ PTSD screenings.
但查尔斯对他们的大张旗鼓持怀疑态度
But Charles is prettyskeptical of such big claims.
我只能告诉你证实这些测验很难
I can only tell you it’s really hard to validate these tests,
而且一个人开始做事先要学会谦逊
and one should start with modest claims
然后在这个基础上才能更进一步
and build on them.
在科学界 巨大的橡树都是从橡子慢慢长成的
In science, great oaks from little acorns grow.
不计后果的研究会将人们的健康置于险境
And reckless research could put people’s health at risk.
因为你的最终结果不是假阳性
Because you’ll end up with false positives
就是假阴性 然后结果就是
and false negatives, so you’ll end up
人们声称可以诊断抑郁症
with people saying they can diagnose depression,
但他们会对一部分人过度诊断
but they’ll over-diagnose it in some people
另一部分人则诊断不足
and under-diagnose it in others.
另一部分专家担心
Other experts are worried about what this means
这对隐私的影响 尤其是当这些小研究
for privacy, especiallyif these small studies
并不那么准确
aren’t that accurate.
假如因为软件错误地将你标识为高风险客户
What if you’re denied amortgage because software
你没能贷款成功会怎么样
mistakenly flags you as a risky customer?
即使分析系统是准确的 又有什么意义?
What does this mean even ifthe systems are accurate?
我们张张嘴巴就放弃了多少隐私权?
What privacy rights are we waiving just by opening our mouths?
我们处在被谷歌 脸书 Siri包围的世界
The world we live in the Google, Facebook, Siri world
在这里 我们只有极其有限的隐私
we live in is one in which we have very limited privacy,
当然无耻的经营者能通过你的声音推断出
and certainly unscrupulous operators could make inferences
各种各样的事情 诸如
about all kinds of thingslike, are you someone
你是不是想买一辆捷豹敞篷车?
who wants to buy a Jaguarconvertible from your voice?
我又怎么知道
How do I know?
最后 我们将很难忽视摆放在客厅的
Ultimately, it’s hard to ignore those home assistants
这些家居助手
sitting in our living rooms.
日趋普遍 它们将给大公司提供
More and more, they’re providing huge companies
我们无数个小时的语音
with countless hours of our voices.
大型科技公司目前尚未涉及
The big tech companies are not involved
这类操作
in this kind of work now,
但他们坐拥数据宝库
but they’re sitting on a goldmine
所以不难想象未来我们将
of data, so it’s nothard to imagine a future
被收集和分析我们说的每一句话的
where we’re surroundedby devices that capture
各种设备所包围
and analyze our every utterance.
每个人都在猜测他们会用这些信息干什么
What they do with thatinfo is anyone’s guess.
Siri
Siri.
喂 Siri
Hey, Siri.
喂 Siri
Hey, Siri.
好的 苹果
Okay, Apple.
好的
Okay.

发表评论

译制信息
视频概述

高度智能化的未来,我们声音中所包含的信息可能带来的利与弊

听录译者

收集自网络

翻译译者

陈光

审核员

审核员_Y

视频来源

https://www.youtube.com/watch?v=aqxlfpPEmhw

相关推荐