Electrical audio signal processing systems and devices – One-way audio signal program distribution – Public address system
Patent
1988-10-07
1990-08-07
Harkcom, Gary V.
Electrical audio signal processing systems and devices
One-way audio signal program distribution
Public address system
G10L 506
Patent
active
049474367
DESCRIPTION:
BRIEF SUMMARY
BACKGROUND OF THE INVENTION
The present invention relates to apparatus for speaker identification i.e. for rescognising or verifying the identity of a person whose speech forms the input to the apparatus.
SUMMARY OF THE INVENTION
According to the present invention there is provided an apparatus for speaker identification comprising: means for deriving from each of successive sample periods of speech input thereto a digital word representing the frequency spectrum thereof; a store for storing codewords indicating for each digital word the relative probability of the frequency spectrum corresponding to that word occuring in the speech of known speakers which the apparatus has been trained to identify, the digital word output of the deriving means being connected to address inputs of the store; and control means responsive, during input of speech by a speaker to be identified, to codewords read out from the store to output, in accordance with a predetemined criterion, data indicating which, if any, of the known speakers is deemed to have been identified.
In a preferred arrangement, the apparatus is arranged in operation during a training sequence, for each sample period, as follows: the store does not exceed a predetermined number-to-store at that location a codeword identifying the speaker whose speech has given rise to the digital word representing that address. the store exceeds the predetermined number, store at that location a codeword reserved for indication of multiple users.
Thus, the presence of a known speaker's code at any given location indicates that the probability of the corresponding frequency spectrum occuring in the speech of that speaker is high relative to its probability of occurring in the speech of other speakers. The presence of the multi-user code, on the other hand, indicates that the probability of occurrence is not particularly high for any one speaker rather than another.
In practice, prior to initial training, the store will require to be cleared by writing to it a code not used for any other purpose. The presence after training is complete of such a `not used` code in any location indicates that the probability of the corresponding spectrum occuring in the speech of any of the known users is low.
In a preferred arrangement the control means comprises a respective counter means for each known speaker and is arranged in operation to respond to each codeword read out from the store by incrementing the counter corresponding to the known speaker (if any) indicated by that codeword as having a relatively high probability of being associated with the frequency spectrum corresponding to the address from which the codeword was read, and the predetermined criterion is that the speaker who shall be deemed to be recognised shall be the one whose counter has the highest count.
Brief Description of the Drawings
Some embodiments of the insertion will now be described, by way of example, with reference to the accompanying drawings, in which:
FIG. 1 is a block diagram of one form of apparatus according to the invention;
FIG. 2 is a more detailed block diagram of the apparatus of FIG. 1; and
FIGS. 3a, 3b and 3c illustrate for the coders shown in FIG. 2.
Detailed Description of the Preferred Embodiment
In FIG. 1, speech signals from a user are input, for example via a microphone M, to a filter bank F to analyse the frequency spectrum of the input speech. Filter outputs representing respective frequency bands are rectified and smoothed and then sampled (S) and coded (C) to together provide the address input to a read-write memory S. For each sample the contents of the address thus accessed may in general contain one of three types of label:
Before being used for speaker identification, the apparatus must undergo a training sequence in which each authorised user speaks into the microphone. The object is to record in the store which spectral combinations are peculiar to certain users, and which are not. Thus, in the training mode, the following action is taken for each sample. particular us
REFERENCES:
patent: 3737580 (1973-06-01), Poza
patent: 4100370 (1978-07-01), Suzuki et al.
patent: 4344031 (1982-08-01), Kuhn et al.
patent: 4394538 (1983-07-01), Warren et al.
patent: 4720863 (1988-01-01), Li et al.
patent: 4752958 (1988-06-01), Cavazza et al.
patent: 4763278 (1988-08-01), Rajasekaran et al.
patent: 4776017 (1988-10-01), Fujimoto
patent: 4783805 (1988-11-01), Baker et al.
patent: 4797927 (1989-01-01), Schaire
ICASSP 82, IEEE International Conference of Acoustics, Speech and Signal Processing, Paris, May 3-5, 1982, vol. 3 (IEEE, New York, U.S.), H. Ney et al.: "Speaker Recognition Using a Feature Weighting Technique", pp. 1645-1648.
IEEE ASSP Magazine, vol. 3, No. 4, Oct. 1986 (IEEE, New York, U.S.), D. O'Shaughnessy: "Speaker Recognition", pp. 4-17.
ICASSP 83, IEEE International Conference on Acoustics, Speech and Signal Processing, Boston, Apr. 14-16, 1983, vol. 2, (IEEE, New York, U.S.), K. P. Li et al.: "An Approach to Text-Independent Speaker Recognition with Short Utterences", pp. 555-558.
Instruments and Control Systems, vol. 45, No. 19, Oct. 1972 (Radnor, U.S.), "Automatic Voice Recognition", pp. 29-32.
ICASSP 84, IEEE International Conference on Acoustics, Speech and Signal Processing, San Diego, Mar. 19-21, 1984, vol. 2 (IEEE, New York, U.S.) M. Krasner et al.: "Investigation of Test-Independent Speaker Identification Techniques Under Conditions of Variable Data", pp. 18B.5.1-18B.5.4.
Greaves Alan J.
Millar Paul C.
British Telecommunications public limited company
Harkcom Gary V.
Knepper David D.
LandOfFree
Speaker verification using memory address does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Speaker verification using memory address, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Speaker verification using memory address will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-966085