JPS60164798A - Monosyllabic voice recognition equipment - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Field of Application] The present invention relates to a monosyllabic speech recognition device that is particularly simple in structure and yet can improve recognition rate.

[Prior Art] In recent years, voice recognition devices have come into practical use as input devices for computers, various control devices, and the like. Speech recognition devices that can recognize human speech as they are have various advantages, such as requiring no special training for use and no restrictions on line of sight or limbs. However, many speech recognition devices currently in use are word speech recognition devices that perform recognition on a word-by-word basis, and while they have the above-mentioned advantages, they have the disadvantage of having a limited number of words.

For these reasons, monosyllabic speech recognition systems that recognize speech in units of syllables have recently been attracting attention. As is well known, the Japanese language system is composed of phonetic characters, meaning that each syllable corresponds almost one-to-one to each kana, so monosyllabic speech recognition systems can be used as various input devices. . In particular, with the spread of Japanese word processors and workstations, many attempts have been made to use this monosyllabic speech recognition system as an input means for these devices.

By the way, in a monosyllabic speech recognition system, the consonant part is usually extracted from the whole syllable part of the syllable characteristic parameter time series (hereinafter simply referred to as pattern), and the consonant parts are matched. Syllable patterns have fewer characteristics between patterns than word patterns, and the duration of vowel parts is generally much longer than that of consonant parts, so minute differences in similarity between consonant parts can cause fluctuations in the patterns of the following vowel parts. This is because it is masked by

The paper by Mr. Yoshida et al., ``Japanese Monosyllabic Speech Recognition Experiment'' (Proceedings of the Acoustical Society of Japan, 3-2-16, 1979), provides an example of such extraction of consonant parts. In this example, a consonant part is extracted using an algorithm that uses a predetermined position among points where the syllable feature vector changes significantly as a consonant/vowel boundary, and this consonant part is matched with a standard registered consonant pattern using dynamic programming with free endpoints. (hereinafter referred to as DP matching) to obtain consonant information.

In addition, Mr. Furui's paper [Monosyllabic recognition and its application to large word speech recognition] (Proceedings of the Institute of Electronics and Communication Engineers (A), 65)
-A;2. pp175-182.1982) shows another method for cutting out consonant parts. In this example, the beginning of a word is cut out at a predetermined ratio to the total length of the syllable, and this is compared with similar registered beginnings of words to obtain consonant information. For example, performing linear matching.

In addition, the paper by Nakayo et al., “Large language word speech recognition using monosyllable unit input from unspecified speakers (Proceedings of the Institute of Electronics and Communication Engineers)
D), 65-D, 12, pp1558-1565.1
(982) also discloses other extraction methods. In this example, since the subject is an unspecified speaker, the registered patterns can be fixed, and for this reason, the boundary point j between consonants and vowels is determined in advance by visual inspection for each registered pattern. This decision point is, for example, the waveform of the audio signal or the formant 3
− Estimated taking into account factors such as For the unknown input kabataan, DP matching is performed for the registered pattern of the entire syllable and this unknown input kabatatern, and the point i where the optimal path passes through the above-mentioned boundary point j is set as the consonant/vowel boundary point of the unknown input kabataan. Various proposals have been made as methods for obtaining consonant information, including endpoint free DP matching. This example also discloses consonant extraction using the same algorithm as in Furui's paper mentioned above.

Regarding the identification of vowel information, both of the above-mentioned papers adopt a method simpler than DP matching. Then, this vowel information and the above-mentioned consonant information are combined to identify the syllable.

However, in the conventional configuration as described above, it was necessary to employ a complicated algorithm to separate the consonant part and the vowel part. In other cases, fixed registered patterns must be visually inspected in advance, which requires complicated work and deep experience. Furthermore, errors in identifying the separation points between consonants and vowels may increase the number of errors in recognizing syllables, especially their consonant information. In Mr. Furui's example, the consonant part is cut out in a fixed manner based on the ratio to the total syllable length, so there is no problem like the one mentioned above, but since the variation in the consonant/vowel boundary point for each syllable is ignored, it naturally cuts out the consonant part for each syllable. It is thought that the recognition rate is different.

In addition, as a stage for finally recognizing a syllable, there is a method of individually looking at consonant information and vowel information and determining a syllable from a combination of the consonant information and vowel information. For example, the vowel information of "ri" and rkJ
The syllable ``ki'' is identified from the consonant information. Another known method is to use vowel information in advance, select syllables that follow the vowel as recognition candidates, and perform DP matching or the like on these candidate syllables. In the latter case, the articulatory coupling elements between consonants and vowels can be fully considered in the final matching, so that good discrimination can be achieved. Such two-stage evaluation is described in JP-A-54-145409, JP-A-58-52694, and JP-A-58-59498.

However, in this case, it is complicated to perform two-stage evaluation.

As described below, one embodiment of the present invention can solve this problem.

[Problems to be solved by the invention This invention has been made in consideration of the above circumstances,
It is an object of the present invention to provide a monosyllabic speech recognition device that can easily and reliably obtain consonant information without distinguishing consonants from vowels.

[Means for Solving the Problems] In order to achieve the above object, the present invention performs DP matching and distance calculation for unknown input cover turns and registered standard patterns. The consonant information of the unknown Kabataan is identified based on the minimum cumulative distance at a predetermined midpoint between the beginning and end of the word.

In one aspect of the present invention, a minimum cumulative distance is also determined for a second intermediate point that is closer to the end of a word than the intermediate point from which consonant information is obtained, that is, a second intermediate point that contains more vowel information sources, and based on this, vowel information is identified. 7-Consonant information may be identified for the candidate standard pattern in which the identified vowel is the subsequent vowel. That is, two-stage matching may be performed. In this case, the minimum cumulative distance of consonant midpoints can be obtained as a by-product of the distance calculation during vowel information discrimination, and the distance calculation can be completed in one process.

In another aspect of the present invention, when obtaining vowel information, the ending of the word may be known and the DP matching calculation may be performed up to a predetermined midpoint. In this case, the standard pattern of five vowels can reliably identify the vowels of the unknown Kabataan, so
The amount of calculation is significantly reduced compared to when all syllables are referenced.

[Embodiment] An embodiment in which the present invention is applied to a speech recognition device for a specific speaker will be described below with reference to the drawings.

FIG. 1 shows this embodiment as a whole. In FIG. 1, a speaker's voice is supplied to a microphone 1, and this voice is converted into an audio signal and supplied to an A/D converter 2. be done. This A/D converter has, for example, a sampling frequency of 20 KHz and an 8-data bit length of 12 bits. The data from the A/D converter 2 is supplied to the feature parameter extraction section 3, where a feature parameter time series (pattern) is formed based on the above-mentioned data.

In this example, a logarithmized spectrum is used as this characteristic parameter, as will be described in detail later.

Since this example targets a specific speaker, training is performed prior to speech recognition. That is, a speaker utters a predetermined number of syllables to be identified, for example, 68, into the microphone 1, which are sequentially calculated by the A/D converter 2 and the feature parameter extraction section 3, and then sent to the recognition section 4 as each syllable. We will supply standard patterns. In this case, the switching circuit 5 of the recognition section 4 is switched to "a" so that the information is registered in the storage section 6 of the recognition section 4. After this preparatory step, the speaker divides the syllables into 100 syllables, for example, ji.
? As speech is inputted at a speed of i/min, each syllable is derived as an unknown input cover pattern via the feature parameter extraction section 3 and stored in the storage section 6 of the recognition section 4. At this time, the switching circuit 5 is switched to side b. The unknown input cover pattern is sequentially referenced to 68 standard patterns, and the most optimal one among the reference results is output via the output circuit 7 to an output device 8 such as a printer or a monitor. Of course, in addition to the most optimal one, there are also
It may also be possible to output candidates, etc.

The feature parameter extraction unit 3 in FIG. 1 forms a time series of logarithmized spectra as shown in FIGS. 2 to 6. That is, the digital data from the A/D converter 2 is pre-emphasized, and based on this pre-emphasized data, the energy Ei for every time frame i, 10 m5ec in this example, is determined. However, E i = 10 (average of squared values of QogzoC amplitude)
It is. Thereafter, a normalized energy Ejn is obtained from the maximum energy E max and the minimum energy IEmin, and Ej is normalized to a value from 0 to 32, for example.

However, Ei. = 32 (E i −E, m1n) E max
−E min . Then, an appropriate threshold value is set from the time distribution of the normalized energy Ein obtained in this way to determine the boundary between syllables.

On the other hand, short-term spectral analysis is also performed on the pre-emphasized digital data. In other words, while moving the digital data every 10m5ec per frame,
Rhinoblad's fast Fourier transform using a Hamming window time function is performed in the range of m5ec (400 points). The power spectrum obtained in this way is logarithmized, and the spectrum from 1.01lz to 7900Hz is
It is divided into 9 frequency bands. Specifically, 1.0
'OHz and 200Hz form one band, similarly 3001 (z and 400Hz,...,7
00011z to 7900Hz form bands, respectively.

After this, the average value of the power spectrum of the unvoiced part is used as background noise for the voiced part.

(each syllable) subtracted from the power spectrum.

The power spectrum from which background noise has been subtracted, that is, the characteristic pattern, is normalized in the time direction and is further subjected to nonlinear transformation of frequency components. That is, consider a characteristic pattern of times 1 to np and frequency bands 1 to m as shown in FIG. Here np is the number of frames for each syllable, m is the number of bands, and in this example, m = 19. The logarithmized power spectrum at each time and each band is simply indicated by circles. To perform normalization in the time direction, both the standard pattern (word length TR) and the unknown input cover turn (word length Tz) are converted to a predetermined pattern length (Tn) by linear interpolation, as shown in FIG. The pattern length Tn is determined separately by experiment. For example, it may be selected to be 1.2 times the average standard pattern syllable length.

Nonlinear transformation of frequency components is performed as shown in FIGS. 5 and 6. That is, as shown in Fig. 5, the characteristic pattern after time normalization is expressed as vij, and its maximum energy is expressed as ViHA) (= MAX (Vij). Then, as shown in Fig. 6, vij is expressed according to the formula below. It is converted into a value vjj from 0 to 255.

If Vjj < ViMAx−Vb(7), Vj,j
= 0Vij > VjMAxVb (7) In the case, the naori b is determined separately by experiment. When Vb was changed to 30.40.50, the optimum value was vb=40.

The optimum value of vb is the voice peak relative to the noise power.

This is thought to be related to the ratio of the signals. This nonlinear conversion can alleviate the adverse effects of noise power.

Next, the recognition unit 4 shown in FIG. 1 will be explained with reference to FIGS. 7 and 8. The recognition unit 4 includes a storage unit 6, a cumulative distance calculation unit 9, and the like.

The storage section 6 stores the input pattern PI and the standard pattern Phi, and the cumulative distance calculation section 9 executes a DP matching calculation between the input pattern 13-P and the standard pattern PRi. However, this cumulative distance calculation section 9 does not require calculation for matching over the entire pattern. It is sufficient that the calculation is performed at least up to time tl=tV (FIG. 7) on the time axis of the input pattern. Here, time tc in FIG. 7 is an intermediate point for obtaining consonant information, and time tv is an intermediate point for obtaining vowel information. This is detailed below.

First, 68 standard patterns PRi are matched for a predetermined unknown cover pattern P. That is, as shown in FIG. 8, a DP matching calculation is performed between the unknown input cover turn PI and the first standard pattern PRi, and at this time, the minimum cumulative distance Dc at tl=tc is determined. Although there are various paths from the starting point (Figure 7, at the beginning of a word) to the grid on tl = tc (the grid within the range indicated by Wc because it is limited to the matching window), the cumulative distance of each of these paths is Take the smallest of these.

After this, further DP matching calculation is continued and t4=
A similar minimum cumulative distance Dv is calculated for tv as well. These minimum cumulative distances Dc and Dv are stored in the storage section 6. Thereafter, a DP matching operation is similarly performed between the unknown input cover pattern and the remaining 67 standard patterns PRi (j, = 2 to 68), and each minimum cumulative distance DC is calculated.
, Dv is recognized.

Minimum cumulative distance Dc for all standard patterns PRi
, Dv are determined, then the vowel is determined from the 68 minimum cumulative distances Dv at t4=tv. That is, the syllable with the minimum cumulative distance Dv is selected, and the vowel of this syllable is set as the detected vowel information.

After this, a syllable in which the vowel of the detection result is the subsequent vowel (for example, if the vowel is "a", it is "a", "ka", "sa", etc.)
...) at ty"tc is selected as candidate syllable data, and among them, the one with the smallest minimum cumulative distance Dc is output as the syllable detection result. The above operation is shown in FIG. Vowel detection unit 10, selection circuit 11
and is executed by the syllable detection unit 12.

The above intermediate points tl:tc, tv are determined by experiment.

For example, tc: about TnX+, tv==Tn×0.
Good results were obtained at about 7.

Table 1 shows the actual results of applying this example to 68 syllables. The number of standard patterns per syllable is one (single template method). In this table, the second candidate is the first candidate.
This is the syllable with the second smallest minimum cumulative distance in the syllable detection unit 12 in the figure.

Table 11 Results of Recognition of 68 Syllables in This Example As is clear from Table 2, this example provides superior recognition results compared to the previous example described above.

Table 28: Comparison of this Example and the Conventional Example As described above, this example does not require segmentation of consonants and vowels, so single syllables can be recognized with an extremely simple configuration. DP when obtaining further vowel information
Consonant information can be obtained by using the by-products of the matching operation, and the amount of calculations can be reduced. Furthermore, the configuration is extremely suitable for hardware implementation.

This embodiment can also improve the recognition rate. This can be considered as follows. As described above, the syllable feature vector includes consonant information at the beginning of the word, and 16- includes vowel information over a wide range from the middle of the word to the end. To illustrate this schematically, the 9th
As shown in the figure. Then, by visual inspection or experience, the boundary between the two is arranged as indicated by a broken line. However, consonants influence the following vowels, and the transition region in particular is thought to contain important elements for determining consonants. Therefore, detecting the boundaries between consonants and vowels to extract consonants actually makes the consonant information uncertain. In this example, the intermediate point tc for obtaining consonant information is determined by experiment without worrying about the consonant/vowel boundary.
For example, since it is determined as shown in FIG. 9, more preferable recognition can be performed. Furthermore, since the end point of the path that can be realized in terms of consonant determination is the lattice group shown by WC in Figure 7, the corresponding end point of the standard pattern can be selected within this range, and delicate matching in the consonant region between the input cover turn and the standard pattern can be achieved. can be executed with a higher degree of freedom, which is thought to improve the recognition rate.

Note that this invention is not limited to the above-described embodiments, and various modifications can be made without departing from the spirit thereof. For example, the present invention may be applied to 101 monosyllables including syllables, and multiple templates may be used. Table 3 shows the results of 101 monosyllable recognition using the same configuration as the above embodiment, and Table 4 shows the results of comparison with the previous version using multiple templates.

Table 3. Recognition results of this invention for 1.01 syllables 1
0- Table 4. Recognition Results of Multiple TemplatesAlso, in the above-described embodiment, DP matching is performed from the beginning of a word when obtaining vowel information, but when obtaining vowel information, the ending of a word may be known and DP matching may be performed. In this case, vowels can be determined sufficiently by setting 20 to 30% of the word length as the midpoint. There are only five standard vowel patterns: ``a'', ``i'', ``tsu'', ``ko'', and ``o''. This is because recognition errors are less likely to occur due to the influence of consonants, and unlike the above-described embodiments, there is no need to obtain 68 or 101 consonant information as a subsidiary from the calculation for vowel information. Therefore, the amount of calculation can be extremely reduced. Of course, 201 corresponding candidate syllables are narrowed down from the recognition results regarding vowels. Separately, 68 or 101 standard patterns are prepared for identifying consonant information, and reference to consonant information is made for candidate syllables according to the present invention, and identification of input syllables is finally completed.

In addition, in the DP matching calculation when obtaining consonant information, the starting point and intermediate point tc may be configured to be variable, and syllable recognition is performed for each of multiple sets of starting points and intermediate points, and the final method is determined by majority vote. It is also possible to have the syllable determined by the syllable.

Of course, it is also possible to select an intermediate point on the time axis of the registered standard pattern and obtain consonant information based on the minimum cumulative distance at this intermediate point.

[Effects of the Invention] As explained above, according to the present invention, distance calculation is performed by DP matching between unknown input cover patterns and registered standard patterns, and between the beginning and the end of the word of these unknown input cover turns or registered standard patterns. The consonant information of the unknown input kabataan is obtained based on the minimum cumulative distance at a predetermined intermediate point. Therefore, there is no need to discriminate consonant/vowel boundary points, and the configuration can be simplified, and errors associated with determining boundary points can be eliminated. Moreover, since the matching for obtaining consonant information can be given a degree of freedom, the recognition rate can be improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention, and FIGS. 2, 3, 4, 5, and 6 are for explaining the feature parameter extraction unit 3 of the embodiment in FIG. 1. , FIG. 7, and FIG. 8 are diagrams for explaining the recognition section 4 of the embodiment in FIG. 1, and FIG. 9 is a diagram for explaining the effects of the embodiment in FIG. 1. ]...Microphone, 2...A/D converter, 3
...Feature parameter extraction unit, 4...Recognition unit, 8
...Output device. Frequency - I12 number. ■i MAX Input and turn Pl 1 (time direction → Figure 0 procedure correction writing type) % type % ■, Display of the incident 1981 Patent application No. 1.7263 2, Name of the invention Monosyllabic speech recognition device 3. Relationship with the case of the person making the amendment Patent applicant 4, sub-agent April 24, 1980 6. Full text of the specification to be amended 7. Details of the amendment as shown in the attached sheet

Claims (1)

[Claims] Monosyllabic speech that recognizes the unknown input monosyllable by analyzing the speech signal of the unknown input monosyllable and comparing the input feature parameter time series extracted from this speech signal with the registered feature parameter time series registered in advance. In the recognition system,
From the beginning of the word of the input feature parameter time series and the registered feature parameter time series, distance calculation is performed according to the dynamic programming matching method. A monosyllable speech recognition device characterized in that consonant information of the unknown input monosyllable is obtained based on a minimum cumulative distance at a predetermined midpoint between the unknown input monosyllables.