JPH0442679B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to an audio editing device used for audio storage services and the like.

(Prior Art and its Problems) Conventionally, in audio storage services, processing such as compression of the amount of data through encoding has been performed, but no processing has been performed on the content of the audio. Therefore, it has been impossible to extract and store only keywords from input speech. For example, the Institute of Electronics and Communication Engineers technical research report
The paper published in SE79-98 titled ``Sound Storage Service Concept'' mentions the storage and playback of audio, but does not go into any detailed processing of the content of the audio.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a speech editing device having a keyword extraction function that detects accent phrase boundaries in input speech and selects and stores only emphasized accent phrases. It is in.

(Means for Solving the Problems) In order to solve the above-mentioned problems, an audio editing device provided by the present invention includes pitch extraction means for extracting a time series of pitch frequencies of input audio, and pitch extraction means for extracting a time series of pitch frequencies of input audio. a waveform temporary memory for temporarily storing the pitch frequency, means for detecting the boundary of the accent phrase of the input speech from the time series of the pitch frequencies, means for calculating the degree of emphasis of the accent phrase, and the degree of emphasis of the accent phrase. and means for selecting an accent phrase based on the magnitude of the accent phrase and outputting a speech waveform corresponding to the selected accent phrase from the waveform temporary memory.

(Principle of the invention) The principle of the invention will now be explained. Japanese speech is composed of accent phrases, which are groups of utterances that have one accent nucleus.
The audio editing device according to the present invention estimates the accent phrase boundary from the interval where the pitch frequency is a valley, and selects the accent phrase if the estimated average pitch level of the accent phrase is sufficiently large. It extracts, stores, and outputs only accented phrases. Hereinafter, the principle of the present invention will be explained based on an example.

Figure 2 is ``I've been losing money ever since I was a child.''
This is an example showing the change in pitch frequency of the utterance over time. The vertical axis is the pitch frequency of the input audio, and the horizontal axis is the time axis. The time series of this pitch frequency is, for example, "Proceedings of the Acoustical Society of Japan"
(October 1977), pages 35-36, by the method described in the paper titled ``A method for optimal selection of pitch periodic sequences''. The part where the locus of the pitch frequency forms a gentle valley is detected as an accent phrase boundary. A gentle valley is, for example, the minimum of Pitch's time series p _(i) , i = 1...N moving average p _(i) = p (i - 1) + 2p (i) + p (i + 1) / 4 Can be obtained as part. Second
Points a and b in the figure are examples of accent phrase boundaries. Furthermore, the dashed-dotted line represents the average pitch level of each accented phrase. For example, if this average pitch level exceeds a certain threshold value θ, that accented phrase is selected.

In this way, in the present invention, the input speech is divided into accent phrases, and only those with sufficiently high pitch levels are selected, stored, and output, so that the parts of the speech that are uttered with emphasis can be An audio editing device that can selectively store and output data and has a data compression effect can be obtained.

(Example) Next, an example of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention. The input audio 1 is held in the waveform temporary memory 5 and is also supplied to the pitch extraction section 2. The pitch extraction unit 2 calculates a time series of pitch frequencies from the input speech 1 supplied, and the accent phrase boundary detection unit 3 calculates a time series of pitch frequencies.
supply to. The accent phrase boundary detection unit 3 detects a valley portion from the time series of pitch frequencies supplied, and outputs the address on the waveform of that portion to the boundary address memory 4. Further, the accent phrase boundary detection section 3 outputs a time series of pitch frequencies divided for each accent phrase to the pitch level calculation section 6. The pitch level calculation unit 6 calculates the average pitch level of each accent phrase. The accent phrase selection section 7 determines whether the average pitch level of each accent phrase supplied from the pitch level calculation section 6 is sufficiently large, and outputs information on whether or not the accent phrase can be selected. From the speech waveform held in the temporary waveform memory 5, information on the address on the waveform of the accent phrase boundary supplied from the boundary address memory 4 and information on whether or not the accent phrase can be selected supplied from the accent phrase selection section 7 are obtained. Based on , only the waveform corresponding to the selected accent phrase is stored in waveform memory 1.
Output to 0. In response to the input of the output command 8, the control circuit 9 outputs the speech waveform from the waveform memory to the boundary processing unit 11 for each accented phrase. The boundary processing unit 11 adds a pause section as appropriate to the supplied speech waveform of each accent phrase, and outputs the speech output 12.

(Effects of the Invention) According to the present invention, there is provided a speech editing device having a keyword extraction function that detects accent phrase boundaries in input speech and selects and stores only emphasized accent phrases. can get.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention;
FIG. 2 is a diagram showing an example of a time series of pitch frequencies for explaining the principle of the present invention. 2... Pitch extraction unit, 3... Accent phrase boundary detection unit, 4... Boundary address memory, 5... Waveform temporary memory, 6... Pitch level calculation unit, 7...
Accent phrase selection section, 9... Control circuit, 10...
Waveform memory, 11... boundary processing section.

Claims (1)

[Claims] 1. Pitch extraction means for extracting a time series of pitch frequencies of input speech, a temporary waveform memory for temporarily storing the input speech, and detecting boundaries of accent phrases of the input speech from the time series of pitch frequencies. means for calculating the degree of emphasis of the accent phrase; and means for selecting an accent phrase based on the degree of emphasis of the accent phrase, and outputting a speech waveform corresponding to the selected accent phrase from the waveform temporary memory. An audio editing device comprising means.