AN IMPROVED AUDIO
firstname.lastname@example.org Computer Science Dept. Carnegie Mellon University SOAPS 2008, Pittsburgh, PA
WHAT ARE CAPTCHAS?
are tests generated by computers and generally passable by humans but not current computer programs.
THE PROBLEM WITH CURRENT AUDIO CAPTCHAS
some cases the human passing rate is only 70%! To make the CAPTCHAs secure.
. noise was injected into the audio files making it harder for both computers and humans to pass.
Since the current audio CAPTCHAs use a limited vocabulary. it was possible for us to collect enough data to train a system that could pass the current audio CAPTCHAs more than 45% of the time.
.ARE CURRENT AUDIO CAPTCHAS SECURE?
A CAPTCHA is
considered broken once a program can pass it 5% of the time.
and digg Collected 1000 CAPTCHAs per type of audio CAPTCHA to use for training and testing Created an ASR system using machine learning techniques
.HOW DID WE TEST THE CURRENT AUDIO CAPTCHAs?
three different types of audio CAPTCHAs: google. reCAPTCHA.
the .wav file of an audio CAPTCHA Segmentation .selecting portions of the audio which most likely are digits/letters Recognition y Extract features from the segment y Classify segment as digit/letter or noise and output the label Stop once a maximum number of segments are classified
QuickTime and a decompressor are needed to see this picture.
.ALGORITHM DETAILS SEGMENTATION
were manually labeled and segmented. we chose the highest energy peaks in the test CAPTCHA and selected fixed size segments roughly centered at the peaks. For testing. We created training segments using this information.
ALGORITHM DETAILS FEATURES
used three popular techniques for extracting features from speech to derive 5 sets of features from the audio. y Mel-frequency cepstral coefficients (MFCC) y Perceptual linear prediction (PLP) y Relative spectral transform with PLP (RASTAPLP)
decision stumps for weak classifiers For each type of audio CAPTCHA we created enough classifiers to label a segment as a digit.ALGORITHM DETAILS . letter. or noise.
ALGORITHM DETAILS .SVM
a single multiclass classifier using all the training segments (from 900 CAPTCHAs)
ALGORITHM DETAILS .k-NN
5 classifiers corresponding to each of the feature sets Used Euclidean distance as our distance metric Cross-validation gives us k=1
CAPTCHA as an audio file Segmentation y Find the highest energy peak. and extract a fixed size segment centered at that peak Recognition y Extract features from segment y Give segment to classifier and obtain label Stop extracting segments once all segments have been labeled or a max solution size is reached.
ANALYSIS OF CURRENT AUDIO CAPTCHAs
Exact Match Rate
three machine learning techniques to 80 70 perform ASR on the 60 CAPTCHAs 50 % 40 y AdaBoost 30 y Support Vector 20 Machines (SVM) 10 y k-Nearest Neighbor 0 (k-NN)
AdaBoost SVM k-NN
y In other words. create an audio reCAPTCHA.THE GOAL
a secure audio CAPTCHA which will be easier for a human to pass and harder for a computer to pass. Equate solving a CAPTCHA with doing some useful work.
millions of words get digitized each day!
.WHAT IS reCAPTCHA?
digitize text on which OCR fails by using the text as its CAPTCHA. Since millions of people solve CAPTCHAs each day.
. Will help transcribe digital audio on which ASR systems fail. The audio being used was originally recorded with the intention that it should be easily understood by humans.THE AUDIO RECAPTCHA
advantage of the human ability to understand words through context.
HOW WILL IT WORK?
with a database of phrases with known transcriptions. Give user adjacent phrases to transcribe as the CAPTCHA . Store the rest of the solution as transcription
. Check user solution against the database to determine the result of the test.
He died in a
That was the shot that killed Harry Lime. He died in a Harry Lime he died in a sewer beneath Vienna Harry Lime.
ANALYSIS OF SECURITY
independent recognition is difficult. Open vocabularies make it even more difficult for ASR systems AM broadcasts and .mp3 compression cause the loss of important data needed for automatic analysis
Deploy audio reCAPTCHA through reCAPTCHA site. yet remain secure and not too difficult for humans. Help make knowledge captured in audio available in text form
need to be more accessible.
CMU Dr. Roni Rosenfeld.ACKNOWLEDGEMENTS
Dr. CMU Dr. CMU David Huggins-Daines. CMU Sean Hyde. CMU
. CMU Jiri Simsa.
Luis von Ahn. Manuel Blum.