Bei der Erkennung werden Sprachwerte, die aus Abtastwerten des Sprachsignals gewonnen werden, mit Referenzwerten verglichen, wobei die Wörter eines vorgegebenen Vokabulars jeweils durch eine Folge von Referenzwerten gegeben sind. Die Wörter werden dabei aus Phonemen gemäß einem vorgege­benen Aussprachelexikon vorgegeben, und die Referenzwerte für die Phoneme werden in einer Lernphase bestimmt, wobei jedes Phonem innerhalb eines Wortes aus einer in der Lern­phase bestimmten Anzahl von gleichen Referenzwerten besteht. Um Übergänge zwischen Phonemen anzunähern, kann jedes Phonem auch aus drei Abschnitten jeweils konstanter Referenzwerte bestehen. Durch die vorgegebene Anzahl Referenzwerte je Phonem kann die zeitliche Dauer eines Phonems in einem bestimmten Wort besser nachgebildet werden. Es werden verschiedene Möglichkeiten angegeben, die Referenzwerte und den Abstandswert bei der Erkennung zu bestimmen.

Voice signals are picked up through a microphone (31) and voice sections are detected by a voice section detector (32). The voice signals are then processed with every predetermined time Interval by a frequency analyzer (33) having a predetermined number of channels, and the voice signal portions corresponding to respective channels are quantized. The quantized data obtained from these channels are converted into binary datas (34) to form frames each composed of a series of binary data. Each data of the frame corresponds to a channel and is preferably set to an integer number of times of the operation unit (e.g., 4 bits, 8 bits, etc.) of a computer. When a synthetic frame is to be formed by superposing two or more of such frames, the synthetic frame is divided Into hierarchies so that each bit may be displayed by binary notation in each hierarchy. It is further possible to divide a frame into a plurality of subframes in order to effect preparatory comparison using the subframes.

The present invention relates to a speech recognition system of the type which comprises storage means (10, 11) for storing selected parameters for each of a plurality of words in a vocabulary to be used for recognition of an input item of speech, comparison means (42) for comparing parameters of each unknown word in an input item of speech with the stored parameters, and indication means (12, 46) responsive to the result of the comparison operation for indicating which of the plurality of vocabulary words most closely resembles each unknown input word.

According to the invention the speech recognition system is characterised in that the stored parameters comprise for each vocabulary word a set of labels each representing a feature of the vocabulary word occurring at a respective segmentation point in the vocabulary word and the probability of the feature associated with each label occurring at a segmentation point in a word. Further, the comparison means compares the stored sets of parameters with a set of labels for each unknown input word each representing a feature of the unknown input word occurring at a respective segmentation point in the unknown input word.

A waveform recognition system comprising a plurality of detectors of features comprising the combined presence at a plurality of instants spaced at predetermined intervals relative to each other in time of instantaneous amplitudes each satisfying respectively predetermined constraints; means for assigning a plurality of labels and corresponding confidence measures to each of successive portions of said waveform in dependence on the features detected in said portions and storing each label in a buffer corresponding to the rank of the confidence with which the label is assigned relative to other labels assigned to the same portion of data and means for outputting labels from that buffer containing labels assigned with the highest confidence whose confidence measures are in a predetermined relationship with those of adjacent labels in the same buffer when the confidence measures of labels in other buffers containing labels assigned with confidence measures of lower rank satisfy predetermined conditions.

The method consists in classifying the sounds forming the uttered words into eight phonetic classes plus a possible indication of the presence of diphtongs, starting from an acoustic-phonetic analysis of the sounds themselves.

To recognize the uttered words the sequence of classes found out are analyzed by search-tree algorithms of pattern matching with sequences of classes corresponding to vocabulary words, and possibly by dynamic programming algorithms.

The detected classes are: silence, voiced fricatives unvoiced fricatives, plosives, affricates, nasals, semivowels, vowels.

A device for implementing the method is also described.

A speech recognition method and apparatus in which a trajectory drawn by a plurality of time-sequential acoustic parameters obtained from segmented speech signals in the parameter space thereof time-normalizes the segmented speech signals (2, 17) and this trajectory is matched (25, 6) with a standard time-normalized trajectory, which was previously registered (4), to perform the speech recognition. A silence acoustic parameter (30) can also be added to the time-sequential acoustic parameter produced from the segmented speech signals to obtain the trajectory thereof. The length of the trajectory plotted from the segmented speech signals is determined and is used to select the previously registered standard trajectory (4) to be matched. Therefore, in addition to checking the distance between the'segmented speech trajectory and the registered standard trajectory to obtain a match (25, 6), the lengths of the two trajectories are also taken into account, thereby increasing the recognition ratio.

The pattern matching method and apparatus described herein comprises a step of determining a distance d(m,j) between each feature vector b of a reference pattern and a feature vector a_m of an input pattern at each point (m,j) of the i-th block defined in the form of a parallelogram which has the IL frame width of an input pattern and is inclined with respect to said reference time axis, a step of conducting DP matching calculations on the basis of the distance determined; and a step of setting the boundary value of an (i+1) -th slant (oblique) block at values at such a slant line contacting with an i-th block as is obtained by said DP matching thereby to effect said DP matching calculations. This pattern matching method and apparatus can recognise at a high accuracy and at a high speed an input string of words spoken in compliance with various grammars. Furthermore the DP matching can be processed at a high speed with less transfer of pattern data and with less capacity of pattern buffer memories.

Ein Verfahren zur Bewertung der Ähnlichkeit jeweils zweier digital dargestellter Zahlenfolgen, insbesondere Funktionskurven, bei dem jeweils Bestandteile oder Bitgruppen (AKF) zweier Binärwörter (A, B) mittels Schwellwerten in Beziehung zueinander gesetzt werden, um Distanzwerte (di) zu bilden, aus denen eine Ähnlichkeitsbewertungsgröße ableitbar ist. Zur Durchführung des Verfahrens ist eine kaskadierte Anordnung von PROMs (P₁ ... P₁₆; P₁ ... P_IV; P₁₇) und Addieren (ADD" ADD₂) vorgesehen.

A pattern recognition apparatus is provided including a vector generating unit for generating an input vector representing the characteristics of an unknown input pattern, a dictionary unit which stores a plurality of reference vectors for each category, a similarlity calculating unit which calculates a similarity between the input vector and a plurality of reference vectors for each category, a comparing unit which determines the category to which the input pattern belongs by comparing the similarities derived from the similarity calculating unit, and an additional dictionary generating unit which generates additional reference vectors for a particular need or application.

The additional dictionary generating unit generates an additional reference vector obtained by subtracting the components of at least common reference vectors, in the specified category, from the input vector. The additional reference vector is stored in the storage area within the additional dictionary memory corresponding to a specified category.

This pattern matching system comprises a pattern matching control unit (3) for gathering an L number of time points i of the input pattern into a block, a distance computation unit (1) with an input pattern buffer (11) having its address specified by the signal of said control unit, and a reference pattern buffer (12) for temporarily storing the vector at the j-th frame of the reference pattern so that the input pattem and the reference pattern may be read out from the input pattern buffer and the reference pattern buffer to derive the inter-vector distance . An asymptotic equation computation unit includes a difference memory. The asymptotic equation of a dynamic programming is computed for the time point of each reference pattern and for each time point in one block of the input pattern with reference to the distance under the connecting condition of the dissimilarity G_n(j) at the time point (ii-1)·L for the block number ii of the time point of the input pattern, thereby to derive the dissimilarity G_n(j) at the time point ii-L of the input pattern.

The pattern matching apparatus comprises a reference pattern supplying means for supplying a reference feature sequence pattern containing control operators for controlling branching and/or omission, an input pattern supplying means for supplying an input pattern of input feature sequence, and a distance computing section for computing the distance between the feature of said input pattern and said reference feature sequence pattern. A work memory has addresses adapted to be appointed in accordance with the - time point in said reference feature sequence pattern, and is adapted to store the cumulative distance. A recurrence formula computing section executes a DP matching recurrence formula computation in accordance with a plurality of values read out of the work memory and the distance resulting cumulative distance. The control operator and the position at which the control operator appears is stored in a stack and a stack processing section has a stack control section which, when the control operator is detected, conducts the PUSH/ POP operation of said stack in accordance with the kind of the detected control operator, thereby to control the DP matching recurrence formula computation which is to be conducted in the recurrence formula computing section. This pattern matching apparatus can deal with various possible deformations of the pattern even with reduced capacity of the reference pattern memory. Furthermore, the apparatus is most suited to the continuousty uttered words recognition in synchronism with the input speech.

The pattern matching apparatus comprises means (20) for supplying a reference pattern consisting of a time sequence of features in which an address information is contained, means (40) for supplying an input pattern of time sequence of features, a work memory (50) for storing a plurality of cumulative distances, means (60) for computing a DP matching recurrence formula on the basis of a cumulative distance read out from the work memory (50), and control means (10) for supplying an address signal to the work memory (50) so as to control the reading address of the work memory (50) based on the address information. This pattern matching apparatus carries out a DP matching operation in accordance with various types of deformation of patterns. Furthermore, the apparatus is capable of efficiently DP matching with a reference pattern expressed in a general automation form.