JPS59176794A - Word 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 Technical Field The present invention relates to a word speech recognition device with improved performance for unknown input speech.

Background> In general, a word speech recognition device amplifies the uttered word speech with an amplifier such as a microphone amplifier, and then uses feature parameters that can express the characteristics of the speech in an acoustic analysis section, such as BPF group (Bant filter). power spectrum by a group of Hass filters),
The autocorrelation function is analyzed into the number of zero crossings. After this, the feature extraction unit determines the word section and compresses the feature parameter time series within the word section in the time axis direction from the perspective of reducing the amount of pattern memory and reducing the calculation time during matching.
It is created as an input pattern using a predetermined algorithm.

After this, the introductory pattern is registered in advance in a similar manner (targeting a specific speaker).

Or, it is matched with a standard pattern created by many speakers (targeting unspecified speakers),
The most similar one is output as the recognition result.

In this standard recognition, sounds other than the words to be recognized, especially voices carelessly uttered by the user, coughing, etc., as well as sudden noises caused by ambient noise, may be misrecognized. This is a common inconvenience.

<Purpose of the Invention> The present invention aims to improve such inconveniences by adding time length information for each word to a standard pattern and performing pattern matching processing with word length restrictions. If the discriminative ability for the phoneme of the feature is low, perhaps due to the feature expressive ability, very similar patterns are created even for words with different durations other than the recognition target word, which causes misrecognition.

Reduce this misrecognition.

■ In line with the reduction in standard pattern capacity, if the standard ζ turn length is fixed at a constant length regardless of the difference in word length, performance problems may occur due to partial loss of features or a tendency to follow suit as the word length becomes longer. leading to a decrease in Misrecognition caused by this is reduced.

■ In addition, if the word length variance between the recognition target words registered as standard patterns is large, that is, if there are many target words with different lengths, the input word length will be adjusted within the word length tolerance of each standard pattern as a matching process. It can be limited to a certain standard pattern.

This is a preliminary selection operation that narrows down the standard patterns to be matched to a small number by setting a limit on the word length, and is useful for reducing the matching processing time.

<Example> An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention.

The word sound uttered into the microphone 1 is amplified by the preprocessing unit 2, and is subjected to processing such as pre-emphasis as necessary. If the processing after this stage is digital processing, the audio waveform is converted into a digital signal in this preprocessing section 2.

Subsequently, the feature analysis section 3 analyzes the quantities that can represent the characteristics of the voice, the short-time power spectrum 1 autocorrelation function, and the number of zero crossings.

From this feature time series, the word section extraction unit 4 extracts the feature time series +01 and duration information +02 of the word section. The feature time series lot of the word section is compressed in the time axis direction by the pattern creation unit 5 and created as pattern +03.

Pattern log and duration length 102 are selector switch SW
The pattern is sent to either the standard pattern storage section 6 or the input pattern storage section 7 depending on the movement of I or SW2.

Normally, in a recognition device targeted at a specific speaker, Sl is closed in order to register a recognition target word in advance, and the recognition target word is registered in the standard pattern storage section 6. In the case of recognition processing, close S2,
The input word pattern is stored in the input word storage section 7.

If you want to target unspecified speakers, use this selector switch 5W.
1, S to 2 are not included, and the standard pattern storage section 6 stores in advance feature patterns and standard word duration lengths extracted from words uttered by multiple speakers. In this case, this storage section 6 is composed of a ROM or the like.

During recognition processing, the pattern matching unit 8 compares and matches the pattern stored in the standard pattern storage unit 6 with the current input pattern in the input pattern storage unit 7, and selects the most similar standard pattern. The number etc. are output as the recognition result 104.

The control section 9 controls each section and between each section, and has an external input section 10 that can adjust the permissible range from the outside by calculating from the duration of each word in the standard pattern. The allowable setting code 105 input from the external input unit 10 expresses the allowable width in terms of time or the upper and lower limit ratios (represented by @) with respect to the standard time length.

In FIG. 2, the area enclosed by the dotted line in FIG. 1 is constructed with a 1-bit microcomputer-1. The microcomputer 1 is CPtJI2 ROM+3. RAMI41/, and if the standard pattern is intended for unspecified speakers, the standard pattern will be stored in the ROM]3.

For speaker-specific word speech recognition, boards and LSIs have been developed to reduce costs due to recent advances in digital signal processing technology and LSI technology, but speaker-independent recognition requires large and expensive equipment. It's staying.

In speaker-independent word speech recognition, there is no feature parameter that can be applied to all phonemes that has little variation depending on the speaker. However, when performing rough classification, the number of zero crossings is known as a parameter with relatively few individual differences. As a result of taking the number of zero crossings as a parameter and conducting various studies, we found that if we adopt the short-time constant level crossing analysis method, we can use the non-specific voice recognition bag device without the analog circuit part as described above. can be realized with CMO3I chimp LSI.

Figure 3 shows the definition of the number of short-term constant level crossings.

This trace amount is obtained by counting the number of times the audio waveform V crosses a certain threshold level th for each short time frame (period). By adjusting and setting the threshold level to a value slightly larger than the steady ambient noise, it is possible to distinguish between word speech sections and silent sections.

The feature amount can represent the relative strength of the spectrum of the voice, and FIG. 4 shows an example of the feature time series (when a stop occurs) obtained by constant level crossing number analysis. (a) is a voice waveform, and (b) is a constant level crossing series. As seen in the figure, the value is low for voiced sounds, and for unvoiced sounds, especially fricatives such as ゝゝs I', sH''.

A high value is shown for affricates such as "TH". In the zero-level crossing analysis method, the feature analysis unit 3 extracts the number of fixed-level crossings for each frame as a feature.

In FIG. 4(b), point A is the beginning of a word, point B is the end of a word, and the word section extraction unit 4 separates words based on this. As shown in Figure 2, a 1-chip microcomputer 1
1, this processing is performed inside the 1-chip microcomputer II.

FIG. 5 shows an example of a standard pattern stored in the standard pattern storage unit 6 or the ROM 13 of the 1-chip microcomputer II, in which a feature series X is stored for each word.

The standard word duration Y is memorized. Furthermore, as shown in the figure, information such as a result output code Z to be output when this pattern is selected may be included.

FIG. 6 is a flowchart showing the main part of the recognition procedure, and the dotted line indicates the matching procedure based on the word length tolerance.

After starting, first is the external input section! Allowable setting code from 0 to 10
50 input conditions are seen, and this code is read and stored for upper and lower limit calculations to be described later.

After that, detect the beginning of the input word like point A in Figure 4,
If the beginning of the word is detected, the time to the end of the word at point B in the figure, that is, the duration of the uttered word is measured. Next, an input cover pattern is created based on the feature time series of this word interval, and the input
The data is stored in the turn storage section 7 (RAM 14 of the one-chip microcomputer 11 in FIG. 2). The input cover turn contains the word length information measured above. Then, ma-noting processing is started using the word length tolerance value within the dotted line in the flowchart.

First, the standard word duration length Y of the standard pattern stored in the standard pattern storage section 6 (in FIG. Calculate the upper and lower limits of. This determines whether the input word falls within the allowable value, and if so, calculates the inter-pattern distance from the standard pattern. After calculating the inter-pattern distance, or if the input\word is not within the tolerance value, jump this inter-pattern distance calculation and judge whether the standard pattern is finished. B. The standard word duration length Y of the next standard pattern to be compared and verified is taken and the process of 0 or more is repeated for each word until the standard pattern to be compared and verified is completed.

Here, if the discriminative ability for the phoneme of the feature is low, very similar patterns will be created even for words with different duration lengths other than the recognition target word, but by limiting the word length as described above, Calculation of inter-pattern distances is omitted, reducing misrecognition.

In addition, as the capacity of standard patterns decreases, even if standard patterns are fixed at a constant length regardless of differences in word length, the partial loss of features will increase as the word length increases, resulting in poor performance. However, by using the word length itself as information for comparison and verification with the above-mentioned restrictions, it is possible to reduce misrecognition of this word length.

Furthermore, by setting a limit on the word length, it is possible to perform a preliminary selection operation that narrows down the number of standard patterns to be matched to a small number, such as by omitting the calculation of the distance between patterns, which reduces the matching processing time. Helpful.

After such comparison and verification, the one with the minimum distance from the calculated standard pattern is searched. Here, a predetermined threshold value is further defined, and only those that are equal to or higher than the predetermined value are determined to be valid and the results are output. Anything below a predetermined value is output as a beam jet.

In this way, the allowable setting code 10 is input again from the external input section 10.
Return to reading 3. It is possible to input any allowable setting code +05 from the external input section 10, and if it changes, the changed allowable setting code 105 will be read.This sets the optimum allowable width according to the usage situation and speaker. In actual practice, it has been confirmed that even if a limit is added within the permissible range of, for example, ±20 to 30, the performance for unknown input speech is improved and the effect on the recognition rate is small.

In the above example, a method of searching for similar patterns using the distance between patterns as a measure was shown, but the turn is expressed as a symbolic symbol, and the sequence is given as the transition in the time axis direction. If the recognition result is a recognition result that completely matches this transition sequence using a device that has a pattern creation section, the word length tolerance is calculated after the match is made, and the input word length is It is also easy to construct an algorithm that outputs the number of the matched pattern if it is within the range, and rejects the input word otherwise.

<Effects of the Invention> As described above, according to the present invention, time length information for each word is added to a standard pattern, and pattern matching is performed with the word length limited thereby.
Therefore, it is possible to provide a useful word speech recognition device with improved performance against ambient noise (in actual use, etc.).

In particular, if a constant level crossing number with relatively small speaker variation is adopted as a feature parameter for unspecified speakers, the I-chip L
It can be implemented as an integrated circuit such as SI.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing one embodiment of the present invention, Figs. 2-7 are block diagrams when a 1-chip microcombi (user) is used, Fig. 3 is a diagram showing audio waveforms, and Fig. 4 is a block diagram showing an embodiment of the present invention. The figure shows a comparison of the audio waveform (a) and the feature sequence (b).
FIG. 5 is a memory map of the standard pattern, and FIG. 6 is a flowchart explaining the operation of the main parts. l...Microphone, 2...Preprocessing unit, 3...Feature analysis unit,
4. Word section extraction section, 5. Pattern creation section, 6.
. . . Standard pattern storage section, 7. Input pattern storage section, 8 Pattern matching section, 9. Control section, 10. External input section, 11. L-chip microcomputer, 1
2...CPtJ, 13...ROM, +4...RAM
. X...Feature series, Y...Standard word duration length. Agent Patent attorney Aihiko Fukushi (and 2 others), 4 Figure 1 5th Flash

Claims (1)

[Claims] 1. Means for storing standard word duration information in addition to feature series patterns as a standard pattern to be compared and verified; means for measuring the duration of a word; means for calculating a tolerance value from each standard word duration information in the standard pattern; and means for determining whether the tolerance value includes the duration of the input word. and means for performing pattern matching processing only on standard patterns that are within an allowable value according to the determination result. 2. The word speech recognition device according to claim 1, wherein the permissible value is set externally based on the word duration length within the standard pattern. 3 Using a constant level crossing waveform as the voice feature, 1
The word speech recognition device according to claim 1, characterized in that the word speech recognition device is integrated into a chip.