toward speech generated cryptographic keys on resource constrained devices n.
Download
Skip this Video
Loading SlideShow in 5 Seconds..
Toward Speech-Generated Cryptographic Keys on Resource Constrained Devices PowerPoint Presentation
Download Presentation
Toward Speech-Generated Cryptographic Keys on Resource Constrained Devices

Loading in 2 Seconds...

play fullscreen
1 / 21

Toward Speech-Generated Cryptographic Keys on Resource Constrained Devices - PowerPoint PPT Presentation


  • 80 Views
  • Uploaded on

Toward Speech-Generated Cryptographic Keys on Resource Constrained Devices. Fabian Monrose Michael K. Reitery Qi Li Daniel P. Lopresti Chilin Shih. Presented by: Li Meixuan, Li Qihua. Outline . Introduction Background Basic Idea Front-end Signal Processing Security

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'Toward Speech-Generated Cryptographic Keys on Resource Constrained Devices' - paige


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
toward speech generated cryptographic keys on resource constrained devices

Toward Speech-Generated Cryptographic Keys on Resource Constrained Devices

Fabian Monrose Michael K. Reitery Qi Li

Daniel P. Lopresti Chilin Shih

Presented by:

Li Meixuan, Li Qihua

outline
Outline
  • Introduction
  • Background
  • Basic Idea
  • Front-end Signal Processing
  • Security
  • Empirical Results
  • Conclusion
introduction
Introduction
  • Voice is a leading contender for the dominant user input medium in futuristic computing devices.
  • Cryptographic keys to perform encryption will derive from voice input of the user.
  • Implementation of an approach to derive a reputable cryptographic key from spoken user input, in which the entropy of the key is drawn from both the passphrase that is spoken and the speech patterns of the user while speaking it.
background
Background
  • Two main stages in generating cryptographic keys from biometric measurements
  • 1st stage:
    • features (Φ) of raw input are used to compute an m-bit string called feature descriptor
    • Feature descriptors produced by the same user should be ‘sufficiently similar’ while descriptors produced by different users are ‘sufficiently different’
background1
Background
  • 2nd stage:
    • Magnifies the separating property
    • Develops a cryptographic key from the feature descriptor and stored cryptographic data
      • If two descriptors are sufficiently similar, the same cryptographic key will be generated from them
background2
Background
  • Initialization:
    • Generate a cryptographic key K
    • Generate 2m shares of K
      • Aligned in a m x 2 table that is stored on stable storage
  • Upon entry of passphrase:
    • System measures mbiometric features, Φi, of the user’s entry of the passphrase
    • Generates feature descriptorbl(i) determined from the l-th login attempts from the i-th feature Φi(l)
    • bl(i) = 0 if Φi(l)< some fixed threshold value

or

bl(i) = 1 otherwise

    • The system then attempts to reconstruct K using the table elements at positions <i, bl(i) >
background3
Background
  • For each successful login:
    • History of feature descriptor is observed and elements of the table not typically accessed are perturbed randomly.
    • Hence, if b(i)=1, then the <i,0> element of the table is randomly altered.
    • b(i) is a distingushing feature if b(i) is sufficiently consistent that element <i,1-b(i)> in the table is perturbed in this way.
  • The correct user, when inducing feature descriptors consistent with those she has induced in the past, should not encounter any of the altered elements in the table.
  • Security of this technique requires that an adversary who captures the device be unable to efficiently differentiate a random table element from a valid share of K
basic idea
Basic Idea

Dispersing the secret

basic idea1
Basic Idea

Key Reconstruction

basic idea2
Basic Idea

How it works

basic idea3
Basic Idea

How it works

front end signal processing
Front-end Signal Processing
  • The main goal is to translate the sound to digital representation using an analog-to-digital converter
  • The less silence and background noise in the representation after processing, the more consistent the user’s utterances will be, the higher the computational cost of processing
  • The higher the sampling rate, the better the resolution of the reconstructed signal, but more storage is required for saving and processing
front end signal processing1
Front-end Signal Processing

A/DC

Down

sampling

Autocorrelation

analysis

energy

End-point

detection

LPC

analysis

Fames

Voice-only

cepstral

mean

subtraction

Silence

remover

security
Security
  • One potential security weakness is the fact that an adversary who captures the device can conceivably reconstruct the key from not just one element of the table per row, but instead using anym elements of the table
    • It is hence important to have distinguishing features
  • An attacker who captures the device on which the key is generated but who has no information about the user's distinguishing features may attack the system by repeatedly guessing a feature descriptor b at random
    • If there are d distinguishing features then each guess will be successful with probability of 2-d, making it harder to attack the system.
  • Security is improved as m and d/m are increased.
empirical results
Empirical Results
  • To calculate the average number of distinguishing features per user, it is important to define when a feature is distinguishing
  • Let µi and σi denote the mean and standard deviation of feature φi
  • φiis distinguishing if | µi – τi | > k σi
  • k tunes the ‘sensitivity’ of the scheme

=> k must be tuned in order for the user to successfully regenerate his key reliably

empirical results evaluation of ipaq recordings
Empirical Results:Evaluation of IPAQ™ recordings

Figure 1: This graph demonstrates the average number of distinguishing features

per user as a function of k.

empirical results evaluation of ipaq recordings1
Empirical Results:Evaluation of IPAQ™ recordings
  • Gap between the "distinguishing features" and the "true speaker" indicates the number of error corrections needed during the key regeneration process to achieve a reasonably low false reject rate
  • Inverse relationship between security and feasibility
  • Human imposters did not match significantly more than if they had guessed a random feature descriptor
other possible attacks
Other possible attacks
  • Cut-and-paste imposter
    • Concatenate the raw speech samples to yield speech like true user
    • Severe discontinuities at the concatenation boundaries, differences in recording levels
  • Text-to-Speech (TTS) imposter
    • Use traditional TTS signal processing to synthesize the passphrase. Makes use of duration and pitch predictions
    • Predictions may not correspond how the true user speaks, pitch and duration pronounced by user is difficult to reproduce
conclusion
Conclusion
  • The viability of (re)generating strong cryptographic keys from voice remains unproven
  • More extensive trials are needed to fine-tune this scheme
references
References
  • F. Monrose, M. K. Reiter, Q. Li, D. P. Lopresti, and C. Shih. Toward speech-generated cryptographic keys on resource constrained devices. In Proceedings of the 11th USENIX Security Symposium, pages 283–296, August 2002.
  • F. Monrose, M. K. Reiter, Q. Li and S. Wetzel. Cryptographic key generation from voice (extended abstract). In Proceeedings of the 2001 IEEE Symposium on Security and Privacy, May 2001
  • C. Ellison, C. Hall, R. Milbert, and B. Schneier. Protecting secret keys with personal entropy. Future Generation Computer Systems 16:311-318, 2000
  • R. D. Rodman. Computer Speech Technology. Artech House, Norwood, MA, 1999