JPS60154300A - Syllable standard pattern delivering system - 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 an improvement in a method for extracting standard syllable patterns in a voice input device.〇〇〇〈Prior Art〉 Generally speaking, syllable parts are extracted from continuously uttered speech. In the method of registering as a syllable standard pattern, if the syllable portion is inaccurately extracted, an erroneous garden standard pattern will be created, which will have a large effect on recognition performance.

Conventional syllable segmentation methods have the problem that the number of syllable segmentation errors changes as the speech rate changes.This error is due to the fact that the syllable segmentation algorithm is fixed regardless of the speech rate. 0 Purpose> The present invention has been made in view of the above points, and detects the speech rate of registered speech, and selects the detected speech rate from among the syllable boundary candidates output from the syllable boundary detection unit. The purpose of the present invention is to provide a 17Iv extraction method for standard syllable patterns, in which the syllable boundary is first selected, the syllable is cut out, and the syllable boundary is determined together with the operator while the syllable cut section is echoed back by voice. It is said that

<Example> Hereinafter, the invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing the overall configuration of a voice input device embodying the present invention. In FIG. 1, the input registered voice is processed by the voice analysis unit 1 from the voice signal at the input time 1#/c, with power p (t) and spectrum y (t
) etc. are extracted.

The feature parameters extracted in the speech analysis section 1 are input to the speech rate detection section 2, and the speech rate detection section 2
A sound interval and a silent interval are distinguished based on the strength of the power p(1) of the parameter fl inputted by the silent interval detection unit 21 and the sound interval detection unit 22 in the above.

- Also, the average syllable length is calculated by the utterance rate calculation unit 23 in the utterance rate detection unit 2 and output based on the duration of the voiced section of the voice input of the registration sentence whose number of syllables is known. .

That is, when registering all the standard patterns of syllables, the user utters a registration sentence whose number of syllables is known, and the speech rate calculation unit 2
In step 3, the average syllable ML (1/average speech rate) is calculated.

Now, when a sentence containing n syllables is uttered, the duration of the first voiced interval detected by the voiced interval detection unit 22 is '1L (it) (where i21.2...・
+m), the speech rate iQ unit 23 calculates and outputs.

Note that this calculation of the average syllable length may be performed by uttering a registration sentence whose number of syllables is known in advance and calculating the average syllable length L' before manually operating the registration voice for registering the syllable standard pattern. In addition, since the number of syllables is known in advance in the registration text for registering the syllable standard pattern, the average syllable length L'z is calculated during the vocalization operation for registering the syllable standard pattern.
It may be calculated, or furthermore, an average value of the average syllable lengths calculated successively may be calculated and this value may be used when cutting out syllables, which will be described later.

Based on the duration of the silent section detected by the silent section detection section 21, the phrase boundary detection section 3 detects the case where the duration of the silent section exceeds a predetermined length, and detects the silent section. is treated as a bunsetsu boundary and outputs the entire text to that effect.

The syllable boundary detection unit 4 outputs syllable boundary candidates using the special parameters extracted by the speech analysis unit 1, using the speech segmented into phrases by the phrase boundary production unit 3 as a unit. the interval is the syllable length). In this syllable boundary detection section 4.

As shown in Figure 2, it may be difficult to determine whether a syllable boundary exists in a certain time domain 'I''< t < T2, but in such cases, the final determination of the syllable boundary is The selection unit 5 performs this.

The syllable boundary selection unit 5 calculates the syllable length of the syllable boundary candidates detected by the syllable boundary detection unit 4 and the speech rate calculation unit 28.
The syllable boundaries are determined by comparing all the syllables with the average syllable length +Lh.

Now, as shown in Fig. 2, a certain time domain "r, <t<T
In No. 2, it is difficult to determine syllable boundaries, so the syllable boundary detection unit 3 selects several syllable candidate sequences A and B.

Suppose that Ci... is created and output (however, the syllable candidate string A has lengths of a pieces A(1), A(2), -, A
(a) Consists of syllable candidates, syllable candidate strings B, C,...
Similarly, this syllable candidate string A, B, C, .
Average syllable length of: nDA, DB+Dc+...
・is calculated as each.

Here, since the syllables and plosives that come at the beginning of a phrase are often shorter than the average syllable length, it is recognized that 0 < kt < 1, and the syllable at the end of a phrase is often longer, so k2
〉1.

The syllable boundary selection unit 5 selects the text for registration that has the smallest deviation from the average syllable length of 1° among the deviations from the average syllable length DA, DB, DC9, etc. calculated as described above. A syllable candidate string having the same number of syllables as that contained in □ is selected and output as a syllable string.

The speech synthesis section 9 divides the input speech stored in the speech output buffer memory 8 into syllable sections selected by the syllable boundary selection section 5 and echoes them back to the operator for output. The operator determines whether or not the cut out syllable section is appropriate based on the echoed audio output, and if it is appropriate, operates the key human power means 10 to notify the audio registration unit 6 to that effect, and registers the audio. In section 6, standard patterns of syllables are also registered in the syllable standard pattern memory 7 for the cut out syllables.

On the other hand, if the operator determines that the syllable section cut out based on the echoed back audio output is inappropriate, he operates the key manual means 10 to notify the syllable boundary selection section 5 to that effect. In step 5, a syllable candidate string with a small deviation from the average syllable length is selected, and in the same manner, each selected syllable section is divided and echoed back to all operators of the input speech.

This operation is repeated until the syllable section is properly cut out, and the standard pattern of syllables is registered as a voice input to be registered.

Effect> As explained above [7], according to the present invention, the utterance speed of the registered speech is detected, the entire syllable is cut out by selecting on the syllable boundary based on the detected utterance speed, and this syllable is cut out. Since the syllable boundaries are determined by echoing back the section with voice, it is possible to accurately detect and determine syllable boundaries, regardless of differences in the speaking speed of the input voice due to operator characteristics, etc. Not only can the syllable parts be reliably extracted, but the operator can also listen to the echo back of the separated voice for each syllable section to confirm whether or not the syllables have been extracted correctly. , it is possible to prevent the registration of incorrect syllable standard patterns at the voice registration stage.

FIG. 2 is a block diagram showing an example of detected syllable boundaries.

1... Voice analysis section, 2... Speech rate detection section, 21.
... Silent section detection section, 22... Sound section detection section, 23
... Speech rate calculation section, 3. ... Clause boundary detection section, 4.
...Syllable boundary detection unit, 5...Syllable boundary selection unit, 6...
・Voice registration section, 7...Syllable standard pattern memory, 8.
...Buffer memory for audio output, 9...Speech synthesis section,
10...Key human power means.

Agent Patent Attorney Fuku Aihiko (and 2 others) -736- 2nd F, °1

Claims (1)

[Claims] 1. Detect the speech rate of the registered speech, select a syllable boundary from among the syllable boundary candidates detected by the syllable boundary detection unit based on the detected speech rate, and cut out the syllable. A method for cutting out a standard syllable pattern, characterized in that syllable boundaries are determined by echoing back sections by voice.