System and Method for Recognizing Speech Securely

• US 20080172233A1
• Filed: 01/16/2007
• Published: 07/17/2008
• Est. Priority Date: 01/16/2007
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Inventors: Shashanka, Madhusudana, Smaragdis, Paris
• Granted Patent: US 7,937,270 B2
• Time in Patent Office: Days
• US Class Current: 704/256
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/144 Training of HMMs
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: H04K 1/00 Secret communication
• CPC Class Codes: H04L 2209/46 Secure multiparty computation, e.g. millionaire problem
• CPC Class Codes: H04L 2209/50 Oblivious transfer
• CPC Class Codes: H04L 9/008 involving homomorphic encryption

Abstract

A system and method recognizes speech securely. The system includes a client and a server, The client is configured to provide securely speech in a form of an observation sequence of symbols, and the server is configured to provide securely a multiple trained hidden Markov models (HMMs), each trained HMM including a multiple states, a state transition probability distribution and an initial state distribution, and each state including a subset of the observation symbols and an observation symbol probability distribution. The observation symbol probability distributions are modeled by mixtures of Gaussian distributions. Also included are means for determining securely, for each HMM, a likelihood the observation sequence is produced by the states of the HMM, and means for determining a particular symbol with a maximum likelihood of a particular subset of the symbols corresponding to the speech.

Claims

1. A system for recognizing speech securely, comprising:

a client configured to provide securely speech in a form of an observation sequence of symbols;

a server configured to provide securely a plurality of trained hidden Markov models (HMMs), each trained HMM including a plurality of states, a state transition probability distribution and an initial state distribution, and each state including a subset of the observation symbols and an observation symbol probability distribution, and in which the observation symbol probability distributions are modeled by mixtures of Gaussian distributions;

means to determine securely, for each HMM, a likelihood the observation sequence is produced by the states of the HMM; and

means to determine securely a particular symbol with a maximum likelihood of a particular subset of the symbols corresponding to the speech,

2. The system of claim 1, in which the observation sequence of symbols is a feature vector of the speech.

3. The system of claim 1, in which the particular subset of symbols is a transcript of the speech.

4. The system of claim 1, in which the symbols are selected from the group comprising of letters, words, phonemes, syllables, and phrases, or combinations thereof.

5. The system of claim 1, in which for each HMM, the plurality of states, the state transition probability distribution, the initial state distribution, the subset of the observation symbols, and the observation symbol probability distribution form the parameters of the HMM, and the means for determining the likelihood further comprises:

means for securely evaluating the likelihood of each symbol with respect to the observation symbol probability distributions associated with each state.

6. The system of claim 1, in which the secure multi-party maximum index protocol uses the millionaire'"'"'s problem protocol.

7. The system of claim 1, in which for each HMM, the plurality of states, the state transition probability distribution and the initial state distribution, the subset of the observation symbols, and the observation symbol probability distribution form the parameters of the HMM, and further comprising:

means for securely determining a best sequence of the states of each HMM to account for the observation sequence of the symbols.

8. The system of claim 1, in which for each HMM, the plurality of states, the state transition probability distribution and the initial state distribution, the subset of the observation symbols, and the observation symbol probability distribution form the parameters of the HMM, and further comprising;

means for adjusting the parameters of each HMM securely to account for the observation sequence of the symbols.

9. The system of claim 1, in which the speech is only revealed to the client, and HMMs are only revealed to the server.

10. A method for recognizing speech securely, comprising:

providing securely speech in a form of an observation sequence of symbols;

providing securely a plurality of trained hidden Markov models (HMMs), each trained HMM including a plurality of states, a state transition probability distribution and an initial state distribution, and each state including a subset of the observation symbols and an observation symbol probability distribution, and in which the observation symbol probability distributions are modeled by mixtures of Gaussian distributions;

determining securely, for each HMM, a likelihood the observation sequence is produced by the states of the HMM; and

determining securely a particular symbol with a maximum likelihood of a particular subset of the symbols corresponding to the speech.