JPS61238099A - 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 "Field of Industrial Application" The present invention relates to a word speech recognition device based on recognition of phoneme units.

``Prior Art'' Conventionally, in word speech recognition devices based on this type of phoneme unit recognition, the feature parameter time series of input speech is expressed by a series of phonetic symbols in a word dictionary, and the similarity of words is calculated. The recognition result was the one with the highest degree of similarity. In that case, candidate words were selected from the word dictionary using only phonemes.

(For example, Itabashi et al., “Dielectric reduction effect by specifying phoneme sequences in words,” Journal of the Institute of Electronics and Communication Engineers, Vol. J67-D,
1IkL8 (1984-8); Sawai et al., "Human language! Examination of preliminary selection for word speech recognition," Acoustical Society of Japan Speech Study Group Materials, 384-14 (1984-6);) In other words, these methods are partial phonological sequences. This is a selection method that only considers the phoneme order relationship, and does not take into account the connection relationship between -phonemes, that is, whether the phonemes are directly connected or whether there is an unknown phoneme in between. For this reason, the word selection ability was not sufficient and a large number of candidate words were required.

Furthermore, the ability to select words changes depending on the state of the vocalization, such that when the vocalization is clearer, fewer words are selected, and when the vocalization is more ambiguous, the number of words selected is increased. I couldn't give it any consideration. Furthermore, sufficient corrective measures were not taken for phoneme detection errors.

As a result, the number of candidate words for which similarity is determined in the word recognition unit increases, which increases the processing time, and reducing the number of candidate words causes the recognition rate to decrease. According to the present invention, a section in which a phoneme definitely exists is detected from a time series of characteristic parameters of an input speech in units of phonemes, that is, the input speech is segmented in units of phonemes, and a likely part of the speech obtained by the segmentation is detected. (section) phoneme is detected, the same phoneme as the detected phoneme and the connection relationship are selected as a candidate word from the word dictionary section, and only the selected candidate word is similar to the input speech feature parameter time series. Find the degree. In this way, a high recognition rate can be obtained by calculating the similarity with a small number of candidate words.

If necessary, candidate words are selected while correcting phoneme detection errors corresponding to phonemes or phoneme chains.

"Embodiment" The figure shows an embodiment of the invention. The audio input from the input terminal 1 is converted into a digital signal by the feature extraction unit 2, and then subjected to LPG analysis.
milliseconds) are converted into feature parameters. These feature parameters include normalized logarithmic power of the input speech, level and spectral distance from noise, power dip (second derivative of quadratic curve), short-term (e.g. 16 ms) spectral changes, and long-term (e.g. 16 ms) spectral changes. For example, 48 milliseconds) Spectral change 9 Frequency low-to-high power ratio, 5 Vowels and a standard pattern of pellicle (N) to WLR scale value (Spectral distance closeness scale (11), etc.).

The feature parameter time series of the converted input speech is input to the segmentation unit 3, and a section in which segmentation can be reliably performed on a phoneme basis, that is, a section in which a phoneme is definitely present, is detected. This interval in which segmentation can be reliably performed is determined by setting thresholds for each of multiple, preferably three or more feature parameters, such as power, power dip, and short-term spectral change, and using one set of thresholds. Then, segmentation is performed on the characteristic harameter of the human voice, that is, it is determined whether or not the thresholds are exceeded at the same time, and segmentation is performed using another set of thresholds, and the section in which both thresholds are exceeded, In other words, a segment in which the difference between the two segmentation sections is small, for example, 2.3 frames or less, is determined as a section in which segmentation can be reliably performed.

For the sections that have been reliably segmented in this way, the probable phoneme recognition unit 4 detects phonemes with respect to the input speech feature parameter time series.

This phoneme detection can be performed in the same manner as in the past by determining the degree of similarity with a standard phoneme that has the same feature parameter sequence. In this example, it is also detected whether the detected probable phoneme is at the beginning or end of a word, based on the power of the voice and its duration.

The candidate word selection unit 5 uses the detected probable phonemes, maintains their continuity and order, that is, maintains the connection relationship, and selects words with phonemes with the same connection relationship from the word dictionary unit 6 as candidate words. Select as.

When making this selection, errors in the recognition results of probable phonemes are corrected as necessary. For example, for this the phoneme i
! ! A recognition result correction rule section 7 is provided.

Although it is known to some extent that there is a relationship in phoneme recognition that is prone to errors, this relationship is stored in the phoneme recognition result correction rule section 7 in advance as a correction rule. For example, the following correction rules can be considered.

(a) Errors for continuous vowels, such as AI and A, E
It is easy to make a mistake.

(Yama) Errors for semi-vowels and persistent consonants, (Errors in segmentation of C1 word endings, and phonology that easily disappears, where the final phonology disappears and the previous phonology is mistakenly recognized as the end of the word) are known.

+dl It is known that there are errors in devoicing, and phonemes that are easily devoiced.

When selecting a candidate word from the word dictionary unit 6 using the detected probable phoneme, if there is no corresponding candidate word, the phoneme recognition result correction rule unit 7 is referred to to select a likely error in the detected probable phoneme. After correction, a candidate word is selected from the word dictionary section 6.

The candidate words selected in this manner are sent to the word recognition section. The word recognition unit 8 determines the degree of similarity between the input speech feature parameter time series from the feature extraction unit 2 and each candidate word. This degree of similarity can be determined in the same manner as in conventional methods, and the candidate word with the highest degree of similarity to be determined is outputted from the recognition result output section 9 as a recognition result.

Next, a specific example of word candidates, which is the main part of this invention, will be shown. Now in the word dictionary section 6 1) SAKATA 2) MITAK
A3) TAKE) 10 4) KITAKATA 5)
Assume that the word TAKEDA exists.

When the probable sound H recognition unit 4 detects only one sound IE (case 1), 3
), 5) are selected.

(Case 2) When two consecutive phonemes KA are detected, words 1), 2), and 4) are selected. Conventionally, segmentation was not performed, and therefore continuity of phonemes was not detected, and only the order was taken into consideration.For example, when the sound of word 5) is input, the ED is not detected as a phoneme, and the ED is not detected as a phoneme. If A is detected, the word 5) is also considered as a candidate.

(Case 3) If two consecutive phonemes TA are detected and they are word-final islands, words 1) and 4) are selected. Conventionally, segmentation is not performed, so if KA does not clearly appear after TA, the word 2) is also considered as a candidate.

(Case 4) If four consecutive phonemes such as TAKA are detected, words 2) and 4) are selected. Conventionally, the word 5) may also be selected for the same reason as mentioned above.

(Case 5) When two consecutive old phonemes and one phoneme K are detected in that order, the word 2) is selected.

(Case 6) When the phoneme T and the phoneme A, which is not continuous, are detected in that order, 2).

Words 4) and 5) are selected. Conventionally, segmentation is not performed and only the order is looked at, so 1), 2)
, 3), 4), and 5).

(Case 7) When the word dictionary section 6 has YA and MA, and the detected probable phonemes are continuous VAN, V
While selecting a word containing AN from the word dictionary section 6, the word cannot be selected because there is no corresponding word, so the user refers to the correction rule section 7, corrects VAN to YAM, and selects the word YAMA as a candidate.

"Effects of the Invention" As explained above, according to the present invention, segmentation is performed to recognize probable phonemes, and this is used to preliminarily select word candidates. can be reduced, and the recognition processing time can be reduced.

For example, top-down and bottom-up speech recognition systems (Matsunaga et al., Top-Down processing and Bo
"Speech Recognition Combining t toIm-υp Processing" Acoustical Society of Japan Speech Research Group Material 383-49 (1983-12)
) is used in the word recognition unit 8, and when the word dictionary unit 6 with 100 city names is used for audio data of 100 city names uttered by 50 people, the recognition rate is 95.5%, which is higher than the conventional technology. In contrast, the number of candidate words can be reduced to an average of 21.1% and the processing time to 62.8%, respectively, and when 643 city names are used, the recognition rate is 82.0%, and the average number of candidate words can be reduced compared to the conventional technology. The processing time was reduced to 17.2% and the processing time was reduced to 53.8%, respectively.

In the above description, each part is generally processed by a dedicated or dual-purpose microprocessor.

[Brief explanation of the drawing]

The figure is a block diagram showing an example of a speech recognition device according to the present invention. 1: Speech signal input terminal, 2: Feature extraction unit, 3: Segmentation unit, 4: Probable phoneme recognition unit, 5: Candidate word selection unit, 6: Word dictionary for voice recognition, 7: Phoneme recognition result correction rule, 8 : word recognition section, 9: recognition result output section.

Claims (1)

[Claims] (1) The input speech is a time series of feature parameters, and the word recognition unit calculates the degree of similarity to that word from the time series of feature parameters and the word expressed as a series of phonetic symbols from the word dictionary. In a word speech recognition device that recognizes words with a high level of recognition, there is provided a means for segmenting the input speech by detecting sections in which phonemes definitely exist in units of phonemes, and identifying which phoneme the segmented section belongs to. and means for selecting a word with the detected phoneme from the word dictionary section and sending it to the word recognition section for determining the degree of similarity with the characteristic parameters of the input speech. Characteristic word speech recognition device.