Arrangement for recognizing sounds
First Claim
1. A preprocessor for sound signals for use in a sound recognition system comprising:
- a. a filter bank having a first input for receiving an electrical input signal which represents a sound pattern, and a plurality of first outputs for presenting a spectral component signal on each first output as filtered out from said electrical input signals and integrated over predetermined fixed time intervals;
b. a plurality of sound vector detectors each having a plurality of window detectors which are provided with second inputs selectively connected to said first outputs for receiving at least one spectral component signal in each of two threshold detectors for comparison thereof with respective predetermined lower and upper threshold values, each window detector having a second output for outputting a binary signal indicating alternatively a "between-thresholds" or an "out-of-thresholds" situation, wherein each sound vector detector furthermore has a combinatorial logic detector for detecting coinciding "between bounds" signals outputted by all its proper window detectors, and thereupon producing a sound vector identifier signal on a third output;
c. storage means having a detector for detecting occurrence of any sound vector identifier signal and thereupon generating a storage location access control signal for storing in a predetermined succession of storage locations sound vector representation signals successively received on a data input connected to all said third outputs;
d. said storage means including a random access memory and an address generator therefor, the output of said detector being coupled to an incrementing input of said address generator, said storage means furthermore having a data encoder for translating any sound vector identifier signal to a corresponding binary code for storage; and
e. each of said pattern detectors having an address associated therewith, said storage means further including a store for storing said addresses of said pattern detectors which produce an output signal at a predetermined time.
0 Assignments
0 Petitions
Accused Products
Abstract
Sounds or speech signals are converted into an electrical sound and broken down into separate spectral components in a filter bank. These components are integrated over a short period of time and produce a short-time spectrum of the signal. The spectral components are applied, via a corresponding number of parallel leads, to a number of pattern detectors which supply an output signal only if the short-time spectrum corresponds to the pattern adjusted in the relevant pattern detector. To this end, each pattern detector includes two threshold value (window) detectors which supply a signal if the applied input signal lies between two adjustable thresholds. Each threshold value detector is connected to a different output of the filter bank for another spectral range and is generally adjusted to different threshold values. The outputs of the threshold value detectors are combined via an AND-element, so that a signal is produced only if all threshold value detectors of the pattern detector are activated. For each sound or for each phoneme in the case of speech, a pattern detector is provided. When series of sounds or signals are recognized, the series of addresses of the pattern detectors which have successively generated an output signal are stored and subsequently applied to the computer for comparison.
-
Citations
5 Claims
-
1. A preprocessor for sound signals for use in a sound recognition system comprising:
-
a. a filter bank having a first input for receiving an electrical input signal which represents a sound pattern, and a plurality of first outputs for presenting a spectral component signal on each first output as filtered out from said electrical input signals and integrated over predetermined fixed time intervals; b. a plurality of sound vector detectors each having a plurality of window detectors which are provided with second inputs selectively connected to said first outputs for receiving at least one spectral component signal in each of two threshold detectors for comparison thereof with respective predetermined lower and upper threshold values, each window detector having a second output for outputting a binary signal indicating alternatively a "between-thresholds" or an "out-of-thresholds" situation, wherein each sound vector detector furthermore has a combinatorial logic detector for detecting coinciding "between bounds" signals outputted by all its proper window detectors, and thereupon producing a sound vector identifier signal on a third output; c. storage means having a detector for detecting occurrence of any sound vector identifier signal and thereupon generating a storage location access control signal for storing in a predetermined succession of storage locations sound vector representation signals successively received on a data input connected to all said third outputs; d. said storage means including a random access memory and an address generator therefor, the output of said detector being coupled to an incrementing input of said address generator, said storage means furthermore having a data encoder for translating any sound vector identifier signal to a corresponding binary code for storage; and e. each of said pattern detectors having an address associated therewith, said storage means further including a store for storing said addresses of said pattern detectors which produce an output signal at a predetermined time. - View Dependent Claims (2, 3, 4, 5)
-
Specification