KR20200085041A - Language rehabilitation based vocal voice evaluation apparatus and method thereof - Google Patents

Abstract

The present invention relates to a device and method for evaluating a spoken voice based on language rehabilitation and, more specifically, to a device and method for evaluating a spoken voice based on language rehabilitation to extract a feature for a plurality of features for pronunciation spoken by a person with language disorder, who has different vocal features with a general person, measuring a distance with standard features in feature space according to a plurality of features, and determining whether the presence or absence of a disorder in speaking and degree of disorder by an average value that a weight of each feature is applied to the distance by the feature.

Description

Speech rehabilitation based vocal voice evaluation apparatus and method thereof

The present invention relates to an apparatus and method for speech rehabilitation based speech rehabilitation, and more specifically, to extract features for a plurality of attributes for pronunciation spoken by a speech impaired person having speech characteristics different from a normal person, and to A speech rehabilitation based speech evaluation apparatus and method for measuring a distance from a standard feature in a property space according to each other and determining the presence or absence of speech disorder by the average value applied to the distance for each property will be.

In general, most existing techniques for processing human voice have been developed for two purposes:

The first goal is to hear the voice and distinguish the speaker, and the second goal is to recognize what you are saying.

The technique for the first goal is a technique for finding the vocal characteristics of each individual and finding the owner of the found vocal pattern, and the technique for the second goal is to remove the individual characteristics and use some remaining speech information This is how to find out if a vocabulary is pronounced.

Currently, these technologies are mainly used in speech recognition research, and the technologies that can develop speech recognition, that is, whether or not there is a language disorder based on the second technology, are also being studied.

However, as described above, the conventional language disorder test technology only examines the disorder due to the inability to accurately express the speech recognition-based technology, and judges the degree of inconsistency in the pronunciation of the vocal pronunciation itself, starting from impossibility of speech recognition. There was a problem in that speech voice could not be evaluated.

Registered Patent Publication No. 10-1804389 (Notice 2017.12.04.)

Accordingly, an object of the present invention is to extract characteristics of a plurality of attributes for pronunciation uttered by a speech impaired person having speech characteristics different from normal people, measure distances from standard characteristics in the attribute space according to the plurality of attributes, It is to provide a speech rehabilitation-based speech and speech evaluation apparatus and method for determining the presence or absence of speech disorder by the average value applied to the distance for each attribute by the weight for each attribute.

The speech rehabilitation based speech rehabilitation apparatus according to the present invention for achieving the above object includes: a standard voice DB storing standard voice characteristic information for each attribute for standard voice and an evaluation reference DB storing evaluation reference level values Storage unit to; An audio processing unit that receives a voice spoken by the talker and outputs voice data; And receiving the voice data, dividing it into a plurality of frames, sampling and normalizing them in units of frames, detecting a plurality of characteristics for each attribute in a normalized frame unit, and storing the detected characteristics for each attribute and the standard voice DB in advance. The Euclidean distance (similarity) between the characteristics of the feature information for each attribute of the corresponding standard voice for the voiced speech is calculated, and after calculating the evaluation value reflecting the calculated Euclidean distance for each attribute in a complex manner, the calculated evaluation value And a control unit for comparing the evaluation reference level value of the evaluation reference DB and evaluating the spoken voice of the speaker.

The control unit may include: a voice signal processing unit for obtaining and outputting voice data through the audio processing unit; A voice feature extraction unit that receives voice data from the voice signal processing unit, divides it into a plurality of frames, samples and normalizes them in frame units, and detects a plurality of attributes for each attribute in a normalized frame unit; A Euclidean distance calculator that calculates a distance (similarity) between the detected feature-specific feature and the feature-specific feature for the standard voice previously stored in the standard voice DB; An evaluation value calculation unit for calculating an evaluation value that reflects the calculated distance for each attribute in a complex manner; And a voice evaluation unit comparing the calculated evaluation value with the evaluation reference level value of the evaluation reference DB to evaluate the spoken voice of the speaker.

The speech feature extracting unit may include a sampling unit that is divided into a plurality of frames and sampled in units of frames to normalize; An energy spectrum obtaining unit performing a short term Fourier transform on a normalized frame to obtain a power spectrum for the voice data corresponding to the spoken voice; A feature extraction unit for each attribute that extracts energy for each attribute in each frequency section for the power spectrum; A log unit for each section that calculates a log value by taking a log of energy for each section of each attribute; And a discrete cosine transform calculation unit for performing a discrete cosine transform on a curve represented by a continuous log value for each section and outputting the discrete cosine transform value as a feature value.

The feature extracting unit for each attribute calculates the energy of the auditory-based attribute for the spoken voice by applying a Mel Scale filter bank to the power spectrum, and adds the calculated energy to extract the MFCC feature extracting unit ; An LPCC feature extraction unit that calculates the energy of a seong-based attribute for a speech voice by applying a linear scale filter bank to the power spectrum, and extracts a second feature by adding the calculated energy; And RASTA-PLP (Relative Spectral), which applies energy to the power spectrum, calculates energy for a speech speech in which background noise is removed by applying a Bark Scale filter bank and a noise filter, and sums the calculated energy to extract a third characteristic. -Perceptual Linear Prediction) It characterized in that it comprises a feature extraction.

The standard voice DB stores standard voice feature information for a predefined vocabulary, and the control unit sequentially outputs one or more of the predefined vocabularies through a speaker of an audio processing unit to a vocabulary output from the speaker. It is characterized by inducing a voice to be spoken.

The vocabulary is characterized in that it is a U-TAP vocabulary.

The evaluation value calculation unit applies the MFCC, LPCC and RASTA-PLP attribute-specific weights (MFCC->w1, LPCC->w2, RASTA-PLP->w3) and Euclidean distances by attribute to Equation 2 below to evaluate the evaluation values. It is characterized by calculating.

[Equation 2]

Here, Feature_similarity _i is the Euclidean distance of each attribute.

The standard voice DB stores standard voice feature information and weights for each vocabulary for a predefined vocabulary, and the controller sequentially outputs one or more of the predefined vocabularies through a speaker of an audio processing unit to enable the talker to After inducing a voice to be output for the output vocabulary, a weight corresponding to the vocabulary of the input voice is applied according to the derived vocabulary.

The speech rehabilitation-based speech evaluation method according to the present invention for achieving the above object includes: a speech acquisition process in which the control unit acquires speech data for the speech uttered by the speaker through the audio processing unit;

A feature detection process for each attribute in which the control unit divides the audio data into a plurality of frames, samples the frames, and normalizes them, and detects a plurality of characteristics for each attribute in the normalized frame unit; A Euclidean distance calculation process in which the control unit calculates a Euclidean distance (similarity) between the characteristics for each attribute detected and the characteristics for each attribute for a standard voice previously stored in the standard voice DB; An evaluation value calculation process in which the control unit calculates an evaluation value that reflects the Euclidean distance for each attribute calculated in a complex manner; And an evaluation process in which the controller compares the calculated evaluation value with the evaluation reference level value of the evaluation reference DB to evaluate the spoken voice of the speaker.

The feature detection process for each attribute includes: a sampling step of dividing into a plurality of frames and sampling and normalizing by frame unit; An energy spectrum acquisition step of performing a short term Fourier transform on a normalized frame to obtain a power spectrum for the speech data corresponding to the spoken speech; A feature extraction step for each attribute that extracts a feature (energy value) for each attribute in each frequency section for the power spectrum; A log value calculation step for each section that calculates a log value by taking a log for each section of each attribute; A log value calculation step for each section that calculates a log value by taking a log for each section of each attribute; And a discrete cosine transform calculation step of performing a discrete cosine transform on a log value for each section of each attribute to calculate a discrete cosine transform value.

In the feature extraction step for each attribute, applying a Mel Scale filter bank to the power spectrum calculates the energy of the auditory-based attribute for the speech, and extracts MFCC features that extract the first feature by adding the calculated energy. step; An LPCC feature extraction step of applying a linear scale filter bank to the power spectrum to calculate the energy of a voice-based attribute for a spoken voice, and summing the calculated energy to extract a second feature; And a RASTA-PLP (Relative Spectral-Packet Level Procedure) applying a Bark Scale filter bank to the power spectrum to calculate the energy of intonation-based attributes for speech, and summing the calculated energy to extract a third characteristic. ) Characterized by including a feature extraction step.

The method further comprises: a vocabulary vocalization induction process in which the controller sequentially outputs at least one of the vocabularies predefined in the standard speech DB through the speaker of the audio processing unit to induce the speaker to vocalize the output vocabulary. It is characterized by including.

The vocabulary is characterized in that it is a U-TAP vocabulary.

The control unit calculates the evaluation value by applying MFCC, LPCC, and RASTA-PLP attribute-specific weights (MFCC->w1, LPCC->w2, RASTA-PLP->w3) and Euclidean distances by attribute to Equation 2 below. It is characterized by.

[Equation 2]

Here, Feature_similarity _i is the Euclidean distance of each attribute. do.

The present invention measures the three attributes of the voice spoken by the talker, measures the distance from the standard speech in the attribute space according to the measured three attributes, and evaluates the talker's speech by applying weights for each attribute. It has the effect of accurately discriminating the degree of disability according to the difference between the talker and the normal person.

The present invention has the effect of preventing the change in the quality of rehabilitation training according to the rehabilitation therapist because it can accurately determine whether or not the degree of disability and the degree of disability, and to provide a high quality voice rehabilitation service at a lower cost.

1 is a view showing the configuration of a speech rehabilitation based speech rehabilitation device according to the present invention.
2 is a flowchart illustrating a speech rehabilitation based speech rehabilitation method according to the present invention.
3 is a flowchart illustrating a speech feature extraction method among speech rehabilitation-based speech evaluation methods according to the present invention.
4 is a diagram showing voice waveforms and characteristics of a disabled child and a normal child according to an embodiment of the present invention.

Hereinafter, the configuration and operation of the speech rehabilitation-based speech evaluation apparatus according to the present invention will be described with reference to the accompanying drawings, and the speech speech evaluation method according to the description will be described in detail.

1 is a view showing the configuration of a speech rehabilitation based speech rehabilitation device according to the present invention.

The speech rehabilitation-based speech evaluation apparatus according to the present invention includes a storage unit 10, a display unit 20, an input unit 30, an audio processing unit 40, and a control unit 50.

The storage unit 10 is a program area for storing a control program for controlling the operation of the speech evaluation apparatus according to the present invention, a temporary area for temporarily storing data generated during the execution of the control program, and data required for the control program And a data area for permanently storing data generated by the control program.

According to the present invention, in the data area, a standard voice DB 11, an evaluation criterion DB 12, and an evaluation DB 13 are configured.

The standard voice DB 11 defines a number of vocabularies, and stores standard voice feature information for each attribute for each attribute for each standard voice of the defined vocabulary. The vocabulary applies the vocabulary used in Urimal Test of Articulation and Phonation (U-TAP) as shown in Table 1 below.

U-TAP vocabulary Pants, button, desk, bag, candy, pencil, car, zoo, mummy, kissy, tiger, elephant, peanut, ears, swing, rabbit, balloon, robot, figure, nail, eyebrow, monster, fight, sparrow, three , Tweet, tree, grasshopper, phone, shawl

The characteristics for each attribute for the standard voice are defined as the average value of the characteristics for each attribute for the vocabulary of the U-TAP of a normal person. The characteristics for each attribute of the standard voice may be defined by age. The properties are MFCC (Mel Frequency Cepstral Coefficient), LPCC (Linear Prediction Cepstrum Coefficient) and RASTA_PLP (Relative Spectral-Perceptual Linear Prediction), which are known as speech recognition methods to be described later, MFCC is auditory (attribute) based, Attribute) based, RASTA_PLP is based on pronunciation (attribute).

In addition, the standard voice DB 11 further stores weight for each vocabulary by attribute.

The evaluation criterion DB 12 stores evaluation criterion level values and evaluation information for each evaluation criterion level value to define a level for the spoken voice.

The evaluation DB 13 stores evaluation information for each user evaluated for any user.

The display unit 20 displays information according to the operation state of the speech rehabilitation based speech rehabilitation device and various information generated during the operation as one or more of graphics, still images, and videos including text and icons.

The input unit 30 is a key input device such as a keyboard having a plurality of keys for inputting functions and information according to the present invention from a user, and is integrally configured on a screen of the display unit 20 and corresponds to a position on the screen being touched. It includes at least one of a touch pad for outputting the location information, a mouse for moving the moving cursor on the screen and outputting the movement information of the moved cursor.

The audio processing unit 40 receives voice (audio) data from the control unit 50 and outputs it as an audible sound through a speaker SPK, and generates a voice signal for a speaker's pronunciation through a microphone (MIC), and voice The signal is converted into audio data and output to the control unit 50.

The control unit 50 includes a voice induction unit 101, a voice signal processing unit 110, a voice feature extraction unit 120, a Euclidean distance calculation unit 130, an evaluation value calculation unit 140, and a voice evaluation unit 150 Thus, the overall operation of the speech rehabilitation-based speech evaluation apparatus according to the present invention is controlled.

Specifically, the speech induction unit 101 monitors whether a pronunciation evaluation event occurs through the input unit 30, and when the pronunciation evaluation event occurs, a voice corresponding to the vocabulary registered in the standard voice DB 11 of the storage unit 10 A vocabulary request message that induces vocalization is output through any one of the display unit 20 and the audio processing unit 40 to induce the talker to vocalize the vocabulary.

The vocal induction unit 101 provides information on the derived vocabulary to the Euclidean distance calculation unit 130 and the evaluation value calculation unit 140.

The voice signal processing unit 110 monitors whether voice data for the vocabulary-derived vocabulary is input from the audio processing unit 40 and outputs the voice data to the voice feature extraction unit 120 when the voice data is input.

The voice feature extraction unit 120 includes a sampling unit 121, an energy spectrum acquisition unit 122, a feature extraction unit 123 for each attribute, a log unit 127, a discrete cosine calculation unit 128, and a discrete cosine coefficient calculation unit The audio data including (129) is received, divided into a plurality of frames, sampled in units of frames, and normalized, and a plurality of attributes for each attribute in the normalized frames is detected.

Specifically, the sampling unit 121 divides the audio data into a plurality of frames, samples the normalized frames, and normalizes the output. The sampling number may be 2000 or the like. If the number of samples per frame is too small, the reliability of frequency analysis decreases. If it is too large, it is important to fit the proper size because many changes are mixed into one frame, so important changes can be missed.

The energy spectrum acquisition unit 122 performs a short term Fourier transform (STFT) on each frame to obtain and output a power spectrum for the speech data.

The feature extraction unit for each attribute 123 applies a filter bank for each attribute to the power spectrum to output energy for each attribute.

The feature extraction unit 123 for each attribute includes a first feature (MFCC) extraction unit 124 that applies a Mel Scale filter bank that extracts energy of the auditory-based first attribute, and a second attribute based on the saints A second feature (LPCC) extraction unit 125 that applies a linear scale filter bank for extracting energy of the bark and a bark scale filter bank that extracts energy of a third attribute based on accent It includes a third feature (RASTA-PLP) extraction unit 126.

The log unit 127 calculates and outputs a log value obtained by taking a log of energy for each section of each attribute output from the feature extraction unit 123 for each attribute.

The discrete cosine transform (DCT) calculator 128 converts a curve represented by log values for each section of each attribute into an appropriate number of cosine functions having discrete cosine function coefficients, and a discrete cosine transform of each of the appropriate number of cosine functions. Find the coefficient, apply the found discrete cosine transform coefficient, calculate the discrete cosine transform value for each attribute, and output it as a feature (value) for each attribute. The number of cosine functions is preferably 8, but is not limited thereto. The discrete cosine transform coefficients and the discrete cosine transform values themselves are well-known techniques, and detailed descriptions thereof will be omitted.

The Euclidean distance calculator 130 is a feature for each attribute output from the voice feature extractor 120, that is, a discrete cosine coefficient calculator 129, and an attribute corresponding to the vocabulary for the voice data of the standard voice DB 11 The distance between the star features is calculated and output.

The Euclidean distance of each attribute may be defined as in Equation 1 below.

은 표준음성 DB(11)에 미리 저장되어 있는 해당 어휘의 각 속성에 대한 특징인 이산 코사인 변환값이다.x _n is a discrete cosine transform characteristic of the frames of each attribute,

Is a discrete cosine transform value that is characteristic for each attribute of the corresponding vocabulary previously stored in the standard voice DB 11.

The evaluation value calculation unit 140 calculates an evaluation value (Quality) by applying the Euclidean distance calculated by the Euclidean distance calculation unit 130 to Equation 2 below.

Here, Feature_similarity _i is the Euclidean distance of each attribute.

The voice evaluation unit 150 compares the evaluation value calculated by the evaluation value calculation unit 140 with the evaluation reference level value stored in the evaluation reference DB 12 and displays evaluation information corresponding to the evaluation reference level value It is provided through any one or more of the unit 20 and the audio processing unit 40.

FIG. 2 is a flowchart illustrating a speech rehabilitation based speech rehabilitation method according to the present invention, and is a flowchart illustrating a method of receiving an age of a speaker and performing evaluation by age group.

Referring to FIG. 2, first, the controller 50 checks whether a speech evaluation event is generated (S111).

When the utterance evaluation event occurs, the control unit 50 may request to input the age through one or more of the display unit 20 and the audio processing unit 40 (S113).

When the age is requested, the control unit 50 checks whether the age is input through the input unit 30 (S115), and when the age is input, performs evaluation setting according to the age (S117). The evaluation setting is to set characteristics and evaluation values for each attribute corresponding to the age group to which the age group is input.

When the evaluation setting is completed, the control unit 50 may display the corresponding vocabulary in a predetermined order or a random order among vocabularies registered in the standard voice DB 11 through one or more of the display unit 20 and the audio processing unit 40. It is requested to utter (S119).

After the voice request, the control unit 50 monitors whether the voice data for the voice spoken is obtained through the audio processing unit 40 (S121), and when the voice data is obtained, performs a voice feature extraction process to calculate the evaluation value Then, the evaluation level is evaluated by comparing the calculated evaluation value with the evaluation reference level value stored in the evaluation reference DB 12, and evaluation information corresponding to the evaluation level is output (S125).

3 is a flowchart illustrating a speech feature extraction method among speech rehabilitation based speech evaluation methods according to the present invention, and FIG. 4 is a diagram showing speech waveforms and characteristics of a disabled child and a normal child according to an embodiment of the present invention. It will be described below with reference to FIGS. 3 and 4. 4, 401 is a 6-year-old girl's voice when a "shawl" is spoken (1), power spectrum (Power Spectrum) (2), attribute-specific (RASTA-PLP, LPCC, MFCC) energy (3, 4) , 5), and 402 is the original signal (1), power spectrum (2), energy by attribute (RASTA-PLP, LPCC, MFCC) when a 12-year-old girl has spoken a "shawl" (3, 4, 5), and 403 is the voice signal (1), Power Spectrum (2), and attribute (RASTA-PLP, LPCC, MFCC) when a normal person has spoken "Shawl". ) Energy (3, 4, 5).

When the voice data is acquired, the controller 50 performs a voice pre-processing process (S210).

When the voice preprocessing process is described in detail, the controller 50 divides the voice data into (window) frames (S211), samples the divided frames with a predefined number of samples (S213), and performs spline interpolation. Apply to normalize (S215), and perform a short-term Fourier transform (STFT) on the frame of the normalized speech to calculate the power spectrum (Power Spectrum) for the entire speech data as shown in 401 and 402 of FIG. S217).

When the above-described voice pre-processing process is completed, the controller 50 performs a voice feature extraction process (S220) for extracting characteristics for each attribute from the power spectrum.

When the power feature extraction process is specifically described, the control unit 50 performs filtering by applying a filter bank for each attribute to the power spectrum when the power spectrum is obtained, such as 3, 4 and 5 of 401 and 402 of FIG. 4. Energy for each attribute is calculated (S221).

When energy for each attribute is calculated, the controller 50 takes a log for each section of energy for each attribute and calculates a log value (S223). Here, the section means a frame section.

After calculating the log value, the controller 50 performs discrete cosine transform (DCT) on log values for each attribute to calculate a discrete cosine transform value for each attribute and outputs the discrete cosine transform value as a feature value (S227).

When the discrete cosine transform value, which is the characteristic value for each attribute, is calculated, the control unit 50 applies the characteristic characteristic and the reference characteristic by attribute for the standard voice of the corresponding vocabulary stored in the standard voice DB 11 to Equation (1). To calculate the Euclidean distance for each attribute (S230).

When the Euclidean distance is calculated, the control unit 50 calculates the evaluation value by applying the attribute-specific weight defined for the vocabulary to Equation 2 in the Euclidean distance and the standard voice DB 11 (S240). The larger the evaluation value, the closer it is to the normal person, and the smaller the evaluation value, the greater the likelihood of having a speech disorder.

Euclidean distance and evaluation value (Quality) for 401 in FIG. 4 are calculated as shown in Table 2 below, and Euclidean distance and evaluation value for 402 in FIG. 4 are calculated as shown in Table 3 below.

6-year-old girl LPCC MFCC RASTA-PLP Quality 20.75 1190.4 15.45 0.045

12-year-old girl LPCC MFCC RASTA-PLP Quality 13.23 716.5 11.08 0.066

As shown in Table 2 and Table 3, it can be seen that the evaluation value of the 12-year-old girl is 0.066, which is higher than the 6-year-old girl. Therefore, a 6-year-old girl will have a low evaluation level and will likely have a language disorder.

On the other hand, the present invention is not limited to only the typical preferred embodiments described above, but can be carried out by improving, changing, replacing or adding in various ways without departing from the gist of the present invention. Anyone who has a will easily understand. If the implementation by such improvement, modification, replacement, or addition falls within the scope of the appended claims, the technical idea should also be regarded as belonging to the present invention.

10: storage unit 11: standard voice DB
12: Evaluation criteria DB 13: Evaluation DB
20: display unit 30: input unit
40: audio processing unit 50: control unit
101: speech induction unit 110: voice signal processing unit
120: voice feature extraction unit 121: sampling unit
122: energy spectrum acquisition unit 123: feature extraction unit by attribute
124: first feature (MFCC) extraction unit 125: second feature (LPCC) extraction unit
126: third feature (RASTA-PLP) extraction unit 127: log unit
128: discrete cosine transform
130: Euclidean distance calculation unit 140: Evaluation value calculation unit
150: voice evaluation unit

Claims (14)

A storage unit including a standard voice DB for storing standard voice feature information for each attribute for the standard voice and an evaluation criterion DB for storing evaluation reference level values;
An audio processing unit that receives a voice spoken by the talker and outputs voice data; And
It receives the voice data, divides it into a number of frames, normalizes by sampling in frame units, detects a plurality of attributes for each attribute in a normalized frame unit, and detects the characteristics for each attribute and the standard voice DB stored in advance. Calculate the Euclidean distance (similarity) between the features of the feature information for each attribute of the corresponding standard voice for the spoken voice, calculate the evaluation value that reflects the calculated Euclidean distance for each attribute, and calculate the evaluation value and And a control unit for evaluating the spoken voice of the speaker by comparing the evaluation reference level value of the evaluation criteria DB.
According to claim 1,
The control unit,
A voice signal processor for obtaining and outputting voice data through the audio processor;
A voice feature extraction unit receiving voice data from the voice signal processing unit, dividing it into a plurality of frames, sampling and normalizing them in frame units, and detecting a plurality of characteristics for each attribute in a normalized frame unit;
A Euclidean distance calculator that calculates a distance (similarity) between the detected feature-specific feature and the feature-specific feature for the standard voice previously stored in the standard voice DB;
An evaluation value calculation unit for calculating an evaluation value that reflects the calculated distance for each attribute in a complex manner; And
And a speech evaluation unit for evaluating the spoken voice of the talker by comparing the calculated evaluation value with the evaluation reference level value of the evaluation reference DB.
According to claim 2,
The voice feature extraction unit,
A sampling unit that is divided into a plurality of frames and sampled in units of frames to normalize;
An energy spectrum obtaining unit performing a short term Fourier transform on a normalized frame to obtain a power spectrum for the voice data corresponding to the spoken voice;
A feature extraction unit for each attribute that extracts energy for each attribute in each frequency section for the power spectrum;
A log unit for each section that calculates a log value by taking a log of energy for each section of each attribute; And
Speech rehabilitation-based speech evaluation comprising a discrete cosine transform calculation unit for performing discrete cosine transform on a curve represented by continuous log values for each section and outputting discrete cosine transform values as feature values. Device.
According to claim 3,
The feature extraction unit for each attribute,
An MFCC feature extraction unit that calculates the energy of the auditory-based attribute for the spoken voice by applying a Mel Scale filter bank to the power spectrum, and extracts the first characteristic by adding the calculated energy;
An LPCC feature extraction unit that calculates the energy of a seong-based attribute for a speech voice by applying a linear scale filter bank to the power spectrum, and extracts a second feature by adding the calculated energy; And
RASTA-PLP (Relative Spectral-) that calculates the energy for the speech with the background noise removed by applying a Bark Scale filter bank and a noise filter to the power spectrum, and extracts the third characteristic by adding the calculated energy Perceptual Linear Prediction) Speech rehabilitation-based speech evaluation apparatus comprising a feature extraction unit.
According to claim 1,
The standard voice DB,
Stores standard voice feature information for predefined vocabulary,
The control unit,
A speech rehabilitation-based speech evaluation apparatus characterized in that one or more of a predefined vocabulary is sequentially output through a speaker of an audio processing unit to induce the speaker to speak a voice for the output vocabulary.
The method of claim 5,
The vocabulary is speech rehabilitation based speech rehabilitation device, characterized in that the U-TAP (U-TAP) vocabulary.
According to claim 4,
The evaluation value calculation unit,
MFCC, LPCC and RASTA-PLP attribute-specific weights (MFCC->w1, LPCC->w2, RASTA-PLP->w3) and Euclidean distance for each attribute are applied to the following Equation 2 to calculate the evaluation value: Speech rehabilitation device based on language rehabilitation.
[Equation 2]

Here, Feature_similarity _i is the Euclidean distance of each attribute.
The method of claim 7,
The standard voice DB,
Stores standard voice feature information and weights for each vocabulary for a predefined vocabulary,
The control unit,
After one or more of the predefined vocabulary is sequentially output through the speaker of the audio processing unit, the speaker is induced to utter a voice for the output vocabulary, and then a weight corresponding to the vocabulary of the voice input according to the vocabulary derived Speech rehabilitation based speech rehabilitation device, characterized in that applying.
A voice acquiring process in which the controller acquires voice data for the voice uttered by the speaker through the audio processor;
A feature detection process for each attribute in which the control unit divides the audio data into a plurality of frames, samples the frames, and normalizes them, and detects a plurality of attributes for each attribute in the normalized frame unit;
A Euclidean distance calculation process in which the control unit calculates a Euclidean distance (similarity) between the characteristics for each attribute detected and the characteristics for each attribute for a standard voice previously stored in the standard voice DB;
An evaluation value calculation process in which the control unit calculates an evaluation value that reflects the Euclidean distance for each attribute calculated in a complex manner; And
And an evaluation process in which the control unit compares the calculated evaluation value with the evaluation reference level value of the evaluation reference DB to evaluate the spoken voice of the speaker.
The method of claim 9,
The feature detection process for each attribute,
A sampling step of dividing into a plurality of frames and sampling and normalizing by frame unit;
An energy spectrum acquisition step of performing a short term Fourier transform on a normalized frame to obtain a power spectrum for the speech data corresponding to the spoken speech;
A feature extraction step for each attribute that extracts a feature (energy value) for each attribute in each frequency section for the power spectrum;
A log value calculation step by section for calculating a log value by taking a log of the feature of each section of each attribute; And
A speech rehabilitation-based speech evaluation method comprising a discrete cosine transform calculation step of calculating a discrete cosine transform value by performing a discrete cosine transform on a log value for each section of each attribute.
The method of claim 10,
The feature extraction step for each attribute,
An MFCC feature extraction step of applying a Mel Scale filter bank to the power spectrum to calculate the energy of the auditory-based attribute for speech, and summing the calculated energy to extract a first feature;
An LPCC feature extraction step of applying a linear scale filter bank to the power spectrum to calculate the energy of a voice-based attribute for a spoken voice, and summing the calculated energy to extract a second feature; And
A Relative Spectral-Packet Level Procedure (RASTA-PLP) that applies a Bark Scale filter bank to the power spectrum to calculate the energy of intonation-based attributes for speech, and sums the calculated energy to extract a third feature. Speech rehabilitation based speech evaluation method characterized in that it comprises a feature extraction step.
The method of claim 9,
The control unit further comprises a vocabulary speech inducing process of sequentially outputting at least one of the vocabularies predefined in the standard speech DB through the speaker of the audio processing unit to induce the speaker to vocalize the output vocabulary. Speech rehabilitation-based speech evaluation method.
The method of claim 12,
The vocabulary is a speech rehabilitation based speech rehabilitation method, characterized in that the U-TAP vocabulary.
The method of claim 11,
The control unit,
MFCC, LPCC and RASTA-PLP attribute-specific weights (MFCC->w1, LPCC->w2, RASTA-PLP->w3) and Euclidean distance for each attribute are applied to the following Equation 2 to calculate the evaluation value: Speech rehabilitation based speech rehabilitation method.
[Equation 2]

Here, Feature_similarity _i is the Euclidean distance of each attribute.

Images