RU2738660C1 - Diagnostic and rehabilitation method of the patients with voice-speech function disorders - Google Patents

Abstract

FIELD: medicine.

SUBSTANCE: invention refers to medicine, namely to otorhinolaryngology, phoniatrics, and can be used for diagnosing and rehabilitation of patients with pathologies of voice formation. Disclosed is a method comprising collecting voice samples of control patients by pronouncing by control patients various types of test tasks, recording voice samples of control patients in a database, determining frequency-amplitude characteristics of voice by Fourier analysis, determining, by means of a neural network, a mathematical correlation between the obtained sound data and diagnoses of the patients, forming a database of the frequency-amplitude characteristics of the voice versus the voice-speech pathologies and the reference voice of the control patients, recording patient's voice samples by pronouncing said same test tasks, analyzing patient's frequency-amplitude characteristics, determining a patient's diagnosis by comparing frequency-amplitude characteristics of a voice of a patient with voice-speech function disorders with frequency-amplitude characteristics of the voice of control patients from a database and taking into account dependence of diagnosis of control patients on frequency-amplitude characteristics of their voice and rehabilitation. Additionally, the patient's rehabilitation procedures are performed using the exercise complexes including phonopedic gymnastics, with constant monitoring of the dynamics of the process by comparing the characteristics of the patient's voice with the characteristics of his voice, obtained during the initial diagnosis and when performing rehabilitation exercises, as well as recording a reference performance of the exercises and correcting the rehabilitation procedure depending on the results of the rehabilitation dynamics.

EFFECT: method provides accurate diagnosis, high rate of rehabilitation and success of treatment without recurrences, does not require additional expensive equipment, it can be implemented even within the use of modern telephone equipment; method is not expensive and is easy to implement.

1 cl, 1 dwg

Description

The invention relates to medicine, namely to otorhinolaryngology, phoniatrics, and can be used for the diagnosis and rehabilitation of patients with pathologies of voice formation.

Voice disorders are widespread among the population. On the one hand, these are people who broke their voices by accident, for example, by shouting incorrectly. On the other hand, there are patients after surgery on the larynx, aorta, trachea, and thyroid gland. On the third hand, there are voice professionals, where phoniatric pathology occurs in 50% of women and 26% of male teachers (E. Nerriere. Voice disorders and mental health in teachers: cross-sectional nationwide study // BMC Public Health. 2009 Oct. 2. Vol.9. P.370). In a study of 1878 teachers in the Netherlands, it was shown that more than half of the surveyed had voice problems during their working life (FI de Jong. Epidemiology of voice problems in Dutch teachers // Folia.Phoniatr.Logop. 2006. Vol.58, No 3. P .186-198). In the work carried out on a sample of 554 teachers of higher education, it was found that teachers are significantly more likely than workers in other spheres to have a weak or demanding voice, and also more often experience physical discomfort when speaking: 36% versus 1% (E.Smith. Frequency and effects of teacher's voice problems // J. Voice. 1997. Vol.11, No 1. P.81-87). In this occupational group, diseases such as aphonia, polyps and nodules of the vocal folds are more common.

According to I.A. Mikhalevskaya (Mikhalevskaya I.A.Prevention of voice disorders in persons of speech and vocal professions - Candidate of Pedagogical Sciences - M., 2005) and K.K. Yakhin (Yakhin K.K., Galiullina LK Borderline mental disorders in patients with dysphonia (psycho-somatic relationships // Mental disorders in general medicine. 2007. V.2, No. 2. P.17-20) the incidence of the vocal apparatus for teachers is 40%, translators - 31%, kindergarten and nursery teachers - 36%, tour guides - 77% Domestic and foreign authors note an increase in the incidence of the vocal apparatus among teachers from 30-40% in the 30-60s of the XX century to 55-60% of cases - in recent decades. The most susceptible to voice disorders are persons of the vocal-speech professions who do not possess the skills of correct vocal presentation. Incorrect voice-leading technique and, as a consequence, overstrain of the vocal apparatus are common causes of disorders that can be caused by a professional disability.

At the same time, the loss of voice can threaten a person not only with partial or complete disability, but also with communicative and psychological isolation, difficulties in social adaptation and depression (Veltischev D.Yu., Romanenko S.G., Stukalo A.V. Psychopathological problems of voice disorders // Doctor.Ru. No. 4 (63). 2011, pp. 63-69).

Recently, there has been an urgent need for children's phoniatric care due to the high prevalence of voice diseases in children.

The complex nature of voice diseases and rehabilitation requires comprehensive diagnosis and long-term treatment. Correction of voice disorders is an interdisciplinary problem that requires the participation of not only an otorhinolaryngologist, but also a phoniatrist, phonopedist, speech therapist, psychologist and teacher, as well as complex treatment. Among the main directions of treatment, it is worth noting drug treatment, phonopedic correction, physiotherapy treatment, psychological trainings.

So, for example, the method of regulation of functional disorders of the voice in patients with psychoemotional disorders (RU 2273500 C2, 10.04.2006) is based on physiotherapy: the patient is exposed to a pulsed low-frequency electromagnetic field with a pulse repetition rate of 20-52 Hz. With hypertonic dysphonia or with subatrophy of the mucous membrane of the posterior pharyngeal wall, they act remotely through a mirror emitter located at a distance of 20-25 cm from the patient's eyes and contact through remote plates to the region of the cervical spine. With hypotonic dysphonia or with hypertrophic pharyngitis, they act contact through the remote plates, paravertebrally on the lateral surfaces of the neck in the larynx region. The field strength in the therapy zone with remote exposure is 1-2 mV / cm, with contact exposure 2-5 V / cm. The duration of exposure is 3-10 minutes daily, the course of treatment is 10-12 procedures. The method normalizes the neuromuscular tone of the vocal apparatus, promotes regression of neurological symptoms, improves the patient's quality of life.

The method of correcting voice disorders in children and adolescents (RU 2336860 C2, 27.10.2008) is based on the interaction of physiotherapeutic treatment and massage: finger acupressure is performed in a soothing or tonic way 4-5 biologically active points from the series: VC ₂₁ , VC ₂₃ , VC ₂₄ , PC ₃ , PC ₂₀ , TR ₂ , TR ₃ , TR ₆ , TR ₈ , R ₁ , R ₃ , C ₄ , VG ₁₄ , VG ₁₅ , VG ₂₀ , VG ₂₆ , GI ₁ , CI ₃ , GI ₄ , GI ₁₇ , GI ₁₇ , E ₄ , E ₅ , E ₆ , E ₁₁ , E ₁₂ , E ₃₆ , E ₄₄ , VB ₁₂ , VB ₃₅ , VB ₄₄ , IG ₁₉ , MC ₅ , МС ₆ , RP ₆ , P ₇ ... Then, exposure to light is carried out using a phototherapeutic device AFS / K for 5-15 seconds on each of 4 auricular points from the row: 2, 4, 5, 6, 10, 11, 13, 15, 22a, 29, 31, 41, 51, 55, 73, 74, 75, 84, 95, 97, 101, 104. Sessions are held daily, for a course of 10 sessions. The second course is carried out in 10 days. The method increases the effectiveness of treatment by improving adaptation to the loads of the vocal apparatus, increasing the time of maximum phonation, normalizing the tone of the vocal cords and improving the psychological state.

Methods of treatment of patients with persistent functional voice disorders based on a combination of physiotherapy and phonopedic gymnastics are described. In (RU 2413547 C2, 10.03.2011), a method is disclosed, which is carried out by conducting neuromuscular electrophonopedic stimulation, acting on the motor nerves of the larynx with the "vocaSTIM" apparatus with simultaneous performance of vocal exercises, characterized in that during hypotonic dysphonia, neuromuscular electrophonopedic stimulation is performed upper laryngeal nerve, and vocal exercises are performed sequentially: times; once or twice; one two three, etc. up to 10, repeat 5 times, R, R, R ..., repeat 10 times, F, F, F ..., repeat 10 times, W, W, W ..., repeat 5 times, S, S, S ..., repeat 5 times , F, F, F ..., repeat 5 times, X, X, X ..., repeat 5 times, H, H, H ..., repeat 5 times, M, M, M ..., repeat 10 times, pronounce sharply: K, KA, KO, KI, KE, KU, repeat 10 times, pronounce sharply: KRU, KRI, KRA, KRE, KRO, repeat 10 times, RU, BRU, VRU, DRU, TRU, MRU, KRU, repeat 5 times, RI , BRI, VRI, TRI, MRI, DRI, KRI, repeat 5 times.

In (RU 2551218 C1, 20.05.2015), a method of treating patients with persistent functional disorders of the voice is described, which includes conducting electrical stimulation and exercises aimed at restoring the functioning of the neuromuscular apparatus of the larynx. Electrical stimulation is carried out with a current with a bipolar asymmetric pulse shape, with a frequency of 20-120 Hz, a pulse duration of 0.2 ms, with an interval between pulses of 2 seconds. The exposure time is 5-10 minutes. Seven pairs of electrodes are used, arranged symmetrically. The first and second pairs of electrodes are placed in the region of the anterior and posterior abdomens of the digastric muscle on the left and right. The third and fourth pairs of electrodes are placed horizontally on the left and right in the areas of the upper and lower third of the lateral surface of the neck, crossing the sternocleidomastoid muscle. The fifth pair of electrodes is at the level of the 3rd and 6th cervical vertebrae in the back of the neck on the left. The sixth pair of electrodes - at the level of the 3rd and 6th cervical vertebrae in the back of the neck on the right. The seventh pair of electrodes is in the region of the middle third of the trapezius muscle to the left and right of the spine. After that, the patient independently performs a series of exercises. In doing so, he pushes the tongue to the teeth, then pulls the root of the tongue back. Then he inhales and, with his mouth closed, squeezes the wings of the nose for 1 s during one exhalation. After that, without opening his mouth, he inhales and once squeezes the wings of the nose for 5-7 seconds, followed by their opening during one exhalation. Then he inhales, squeezes the wings of the nose and exhales through the mouth, once squeezing the lips for 1 second. Repeats this exercise, increasing the duration of the lips tightening to 5-7 seconds. The patient performs each exercise 3-4 times in a row. During the exercise, the patient bends his head to his chest, pressing his chin against it. After doing the exercises, vibrostimulation is applied to the shoulder girdle area. The impact is carried out with a frequency of 20-60 Hz, an amplitude of 1.5 mm, with a rate of increase in the vibration frequency by 20% in 10 seconds. The exposure time is 10-15 minutes. There are 8-12 daily procedures for the course. The method increases the working range of the voice, restores the phonation time by restoring the contraction of the muscles of the larynx, and normalizes the functions of the respiratory apparatus.

The combination of drug and physiotherapeutic treatment is proposed in a method for correcting the autonomic nervous system in case of functional disorders of the voice (RU 2414935 C2, 03/27/2011), including the effect on the larynx region. The impact is carried out by ultratonotherapy from the device "Ultraton-AMP-2INT". First, they act on the cervical-collar zone, and then on the larynx region. Duration of exposure is from 5 to 10 minutes per area. There are 10-12 procedures per course. With the beginning of ultratonotherapy, Mexidol 125 mg is orally administered twice a day for 2-3 weeks. EFFECT: method provides high clinical efficiency of voice restoration in functional disorders due to normalization of laryngeal nervous regulation, reduces drug intake.

In general, the disadvantages of the methods described in the above sources are the lack of accurate diagnosis and control of treatment without the use of objective methods, in particular, measuring the frequency-amplitude characteristics of the patient's voice.

A known method for predicting and correcting the treatment of functional disorders of the voice (RU 2299009 C1, 20.01.2007). Anamnesis is collected, functional dysphonia is identified and drug treatment, physiotherapy and phonopedic gymnastics are carried out. At the same time, spectral analysis of heart rate variability is additionally carried out, ultra-low-frequency oscillations (VLF) are determined and spectral power density (SPD) is calculated, taking as one point the international unit SPD, equal to 1 s / Hz · 100. Based on the SPM value, the possibility of relapse is predicted, and the necessary medications and physiotherapy are adjusted to restore the voice.

The disadvantage of this method is that the applied method of spectral analysis of heart rate variability is not a direct method for diagnosing voice pathology. The factor of heart rate variability is subject to significant changes due to the influence of other factors, which, accordingly, does not allow to fully diagnose the pathology without the participation of other numerical HRV indicators, which are also subject to assessment within HRV, and not only HF, LF and VLF waves. Thus, the results of examination by this method can be influenced by a number of other factors, not related to voice, with vocal production. A certain disadvantage of the method is also the difficulty of providing reproducible conditions for the patient's emotional state during experiments, in particular, given the high emotional lability, various psychological disorders inherent in persons with pathology of voice formation. This can lead to large errors in determining the SPM, and even to an incorrect diagnosis and, consequently, to an erroneous therapy method. Also, pharmacotherapy is not always advisable; moreover, pharmacotherapy can be more differentiated than only those drugs that are indicated in the patent.

Methods for diagnosing patients with impaired voice functions using spectral analysis of the characteristics of the patient's voice, as well as using deep machine learning, are becoming increasingly important.

So, the method of spectral computer analysis of the voice is known (Vasilenko Yu.S., Meshcherkin A.P., Pavlikhin O.G., Romanenko S.G. Spectral computer analysis of the voice - a method of early and differential diagnosis of violations of the voice function // Electronic resource https : //nikio.ru/spectral-analysis-voice). The methodological recommendations describe a method for assessing the functional state of the vocal apparatus. First, a history of the disease is collected, a general clinical examination, laryngoscopy, microlaryngoscopy are performed. Then the voice is recorded in a special room, which must be provided with sound insulation. Analysis of the stability of the main tone and the duration of phonation of various vowel sounds helps to predict the course of the disease, to assess the degree of probability of the transition of a functional disease to an organic one.

The disadvantage of this method is the lack of a programmed spectrum estimation method, the absence of an artificial intelligence system and machine learning that provide faster and more accurate diagnostics.

A known method for the diagnosis of neurodegenerative diseases using machine learning (KR101936302 B1, 01/08/2019). This patent uses a neural network to diagnose neurodegenerative diseases by analyzing the patient's speech signal. User speech data is converted into image data, which is then injected into a trained neural network and used to determine the diagnostic result.

The disadvantage of this method is a certain loss of specificity of the diagnostic method as a result of converting speech data into pictorial data. Image analysis will allow detecting a limited range of standard pathologies and cannot be used to diagnose complex or combined diseases. In addition, there is a lot of unnecessary data in the spectrum image and, therefore, additional resources will be required for their storage and processing, as a result, training the neural network will take more time and may give worse results.

In our opinion, the closest technical solution to the claimed one is the patent (KR101908955 B1, 10/17/2018). The proposed solution relates to a system for the diagnosis of speech disease, which determines the patient's condition by analyzing his voice. The Voice Disease Diagnostic System detects differences from previously established diagnostic information. The invention includes a program for operating on a smartphone, a module for inputting a voice of a subject, a module for analyzing a subject's condition by comparing a voice of a subject with information about a disease diagnosis based on a set of voices, and a learning module for verifying a voice of a subject with information about a disease diagnosis using a machine learning algorithm ...

The disadvantage of this method is the lack of a patient rehabilitation module. In addition, diagnostics via a smartphone will not give the required accuracy of the diagnosis, because only a narrow range of voice, with distortions, is recorded there. A better microphone is needed. Moreover, there are no special tests in this method.

The objective of the invention is to develop a new comprehensive method for the diagnosis and rehabilitation of patients with impaired voice-speech function using a trained neural network, which provides patients with faster and more accurate diagnostics using specialized test tasks and spectral analysis of the patient's voice, which also provides the possibility of quick and effective rehabilitation in order to accelerate recovery and ensuring the absence of recurrence of diseases, in addition, expanding the arsenal of known methods of the specified appointment.

The technical result of the invention is not only fast and accurate software and hardware diagnostics using the analysis of the frequency-amplitude characteristics of the patient's voice, with high resolution and clarity of indications, but also highly effective assistance in patient rehabilitation through continuous monitoring of the dynamics of rehabilitation and instant adjustments, if necessary, which leads to a rapid restoration of the voice-speech function of patients and the absence of recurrence of diseases.

The problem is solved and the technical result is achieved by the claimed method for diagnostics and rehabilitation of patients with impaired voice-speech function.

The method includes collecting voice samples from control patients by pronouncing various types of special test tasks, recording patient voice samples into a database using a hardware-software complex, and determining the frequency-amplitude characteristics of the voice using the Fourier analysis method. Also, a preliminary history of the disease is collected and patients are examined using clinical techniques: magnetic resonance imaging, laryngoscopy, endoscopy, stroboscopy, narrow-spectrum imaging (NBI) with a diagnosis, which is also entered into the database. Then, the input information for the neural network is prepared, the mathematical correlation between the received sound data and the patient's diagnoses is determined using the neural network, and the neural network is trained. As a result, an extensive database of the dependence of the diagnosis of patients on the frequency-amplitude characteristics of their voice, as well as the reference sounding of the voice of healthy people from the control group, is formed.

Then, using the hardware-software complex, the voice samples of a particular patient are recorded by pronouncing special test tasks. Moreover, the choice of the type of test items is based on the history and the results of preliminary examinations of the patient using clinical techniques. Next, a spectral computer analysis is carried out using the Fourier method of the voice of a particular patient, the frequency-amplitude characteristics are analyzed, with the help of digital filters, regions of characteristic frequencies are distinguished and the amplitude and shape of the peaks of characteristic frequencies are analyzed. Then, using the hardware-software complex, the patient is diagnosed by comparing the frequency-amplitude characteristics of the voice of a particular patient with the frequency-amplitude characteristics of the voices of control patients from the database and taking into account the dependence of the diagnosis of control patients on the frequency-amplitude characteristics of their voice. The diagnosis is confirmed by physical examination. Rehabilitation is carried out by methods known for the diagnosis obtained, while additionally, the patient's rehabilitation procedures are carried out using original sets of exercises (for example, phoniatric gymnastics), selected by a hardware-software complex for specific pathologies, with constant, for example, daily monitoring of the rehabilitation process by an objective assessment of the state of voice characteristics of the patient with the help of a hardware-software complex and correction of the rehabilitation procedure depending on the results of measuring the dynamics of rehabilitation.

The test items for ordinary patients and for voice professionals differ significantly. Test items for voice professionals consist of 12 exercises:

1. Constant tone

2. Maximum phonation length

3. Glissando up

4. Glissando down

5. Whisper-speech

6. Stro-bass

7. Staccato

8. Measurement of evenness

9.Labial vibrant of increasing volume

10.Labial vibrant of decreasing volume

11. Labial vibrant increase in height

12. Lip vibrant of decreasing height

The Speech Patient Test consists of six exercises:

1. Constant tone

2. Maximum phonation length

3. Glissando up

4. Glissando down

5. Whisper-speech

6. Stro-bass

Due to the fact that the structure of the skull, resonator cavities, and the natural tone of the vocal folds of each individual person have individual characteristics, the technique for achieving optimal voice formation will also vary from patient to patient.

As a result of the implementation of the method according to the invention, a correctly delivered speech voice is formed for speech patients, and singers need not only a correctly delivered speech voice, but also a singing voice, which have a great mutual influence on each other in the singer's professional life. The goal of a set of rehabilitation exercises for vocalists is to achieve longer breathing, greater range, and evenness of vibrato.

The claimed method provides for a study of the dependence of various characteristics of the spectrogram of the voice on the type of pathology, the formation and analysis of the voice database. Diagnosis and treatment of diseases of the vocal apparatus are carried out by analyzing the sound spectrum of the patient's voice using Fourier analysis techniques, specially selected digital filters and a deep learning neural network. Fourier analysis is used to preprocess the spectrum of a voice and isolate its frequency characteristics. Digital filters are specially selected and used to sample data characterizing the deviations of the spectrum of a defective voice from the spectrum of a healthy voice, as well as to cut off secondary data that do not affect the diagnosis.

When conducting spectral computer analysis of the voice, the frequency-amplitude characteristic of the voice is usually measured, the Fourier transform is carried out and the main characteristics of the voice are determined: the frequency of the fundamental tone and its stability, the type of voice (soprano, mezzo-soprano, baritone, etc.), frequency and amplitude (dynamic ) voice range, timbre (a set of overtones and their regularity), formant composition of the sound signal ("flight"), vibrato characteristics, phonation duration. The method also allows you to determine the rate of rise and fall of the sound signal ("attack"), the stability of the frequency-amplitude characteristics of the voice ("strength", "evenness", "clarity"), the stability of the characteristics of the voice for a long time 2 - 6 hours (endurance of the voice ), the number of areas of resonant voice amplification - formants, the appearance of additional frequency peaks that are not typical for this type of voice. The difference and advantage of our method is the availability of special tests for recording the voices of vocal patients and "ordinary" patients, the use of digital filters for energy and spectral analysis of the voice. Digital filters are specially designed and used to select data characterizing the deviation of the spectrum of a defective voice from the spectrum of a healthy voice, as well as to cut off secondary data (parts of the spectrum) that do not affect the diagnosis, which also provides significant savings in computing power and speeds up the diagnosis and rehabilitation of the patient. In contrast to earlier works, our technique allows you to characterize the voice on all the points listed above, since professional voice recording is carried out with high resolution in amplitude and time. The use of artificial intelligence allows you to analyze large amounts of data, develop proposals for accurate diagnostics and recommendations for patient rehabilitation.

Computer analysis of the spectral characteristics of the voice has revealed a number of pathologies. With phonasthenia, for example, a sharp decrease or absence of the main tone is observed. With hemorrhage into the vocal fold, there is a decrease in the frequency and amplitude range of the voice, an unstable frequency and timbre of the voice. Hypotonic dysphonia is characterized by a shift in the frequency of the main tone towards low frequencies, a narrowing of the dynamic range, and timbre instability. Thus, computer analysis of changes in the spectral characteristics of the voice and clinical data allows an unambiguous diagnosis of voice pathologies.

FIG. 1 shows an algorithm of the method and a block diagram of a hardware-software complex (PAK) for diagnostics and rehabilitation of patients with impaired voice-speech function.

The hardware-software complex provides the collection and analysis of the database of the comparison group and the group with various pathologies, the study of the dependence of various characteristics of the voice spectrogram on the type of pathology. The complex allows you to investigate and monitor changes in spectrogram indicators during rehabilitation. Analysis of the dynamics of changes in the spectral characteristics of the voice provides, if necessary, the adjustment of the individual rehabilitation program.

The following requirements are imposed on the hardware part of a professional device for spectral analysis of voice: high sound sensitivity, low level of intrinsic noise, linear amplitude-frequency characteristic, large dynamic range.

When implementing the invention, a studio set Focusrite Scarlett Solo Studio 3rd Generation was used as a hardware module of the complex, consisting of a microphone, an external sound card, an amplifier, headphones, microphone and USB cables, and software. The CM25 MkIII large-diaphragm 20mm cardioid condenser microphone delivers studio-quality audio recordings that accurately capture the original voice. The kit also includes a third generation Scarlett mic pre-amplifier, A / D and D / A converters with 24-bit converters and 192 kHz sampling rate. The audio interface is connected to a computer using a USB cable and does not require an external power supply. The Scarlett Studio HP60 MkIII reference closed headphones are used to monitor the voice recording process.

The characteristics of the microphone input, which largely determine the characteristics of the entire hardware part: dynamic range: 111 dB; frequency response: 20 Hz to 20 kHz, ± 0.1 dB; nonlinear distortion factor including noise: <0.0012%; equivalent noise level at the input: –128 dB; maximum input level: +9 dB at minimum gain, gain range: 56 dB.

The software consists of four main parts: for controlling sound recording equipment, for spectral analysis, for collecting voice information and forming a database, for analyzing large amounts of data using "artificial intelligence". The set of equipment includes the software required for voice recording, including two fully functional audio workstations Ableton Live Lite and Pro Tools Firs Focusrite Creative Pack.

For energy and spectral analysis of the voice, the program "Digital Filter" is used - a filter of frequency ranges.

There are four types of programs used in the hardware-software complex.

1) Program for collecting and preliminary data processing. The bulk of the data is occupied by sound samples. Each sound fragment is a few seconds long and takes up to 1 MB. Up to 10 samples are collected from such fragments for each subject. To collect the minimum required amount of data, at least 100 people must be surveyed, so the minimum amount of data is up to 1 GB. In the future, it is possible to increase the number of patients and data, and, accordingly, the volume of the database, for more effective diagnosis.

2) Neural network training program. The program requires large computing resources. To speed up calculations, a powerful video card with at least 8 GB memory is required. The program uses the voice data received by the data collection program. The size of the database used ranges from 1 GB to several hundred gigabytes.

3) Server program. It uses a trained neural network. For its operation, a configuration server is used at least 8 GB of RAM, at least a 16-core processor. The server runs under the operating system of the Linux family. Alternatively, use a virtual cloud server provided by a cloud provider.

4) Client program. This program calls on the server program to perform diagnostics and obtain treatment results. For its operation, a computer is used with a configuration: at least 2GB of RAM, at least a 4-core processor running the operating system Windows 7 and higher.

To process and classify data, we used one of the most relevant methods: artificial intelligence and, in particular, machine learning. Deep learning with the help of so-called "neural networks" has shown good efficiency. We have successfully used neural networks to classify voice spectra. By training a neural network with a large amount of data, characteristic common features are identified that combine similar spectra of voices that have, for example, the same defect. To effectively highlight the desired features, the set of recorded voices must have the same parameters, for example, it was the reproduction of the same special test tasks. Further, the neural network training continues on other input parameters, which makes it possible to improve the classification accuracy. The trained neural network is then used to make a diagnosis. Based on the diagnosis, a set of recommendations and exercises for treatment and rehabilitation is issued.

The neural network does not give one hundred percent predictions, it produces a prediction with a certain degree of probability. Obtaining a forecast over time, we determine the dynamics of the probability change. Thus, it is revealed, for example, how effective the rehabilitation of patients with voice problems was, examining how the probability of coincidence of characteristics with a reference healthy voice changes. This allowed the use of deep machine learning to control the dynamics of patient rehabilitation. During rehabilitation, control recordings of the patient's voice were made, changes in spectrogram indicators were analyzed using a neural network for the purpose of training it. If necessary, this allows for quick adjustments to the individual rehabilitation program.

The present method is complex, but at the same time, emphasis is placed on voice diagnostics and rehabilitation, and no additional expensive equipment is required. Within the framework of telephone equipment, it is also possible to apply a technical solution for speech patients. The method is not costly and easy to implement.

This technical solution differs from the existing developments with a modern highly effective set of exercises for individual rehabilitation, taking into account the world experience of working with the human voice. The set of exercises consists of breathing exercises, lymphatic drainage gymnastics, stretching exercises for the muscles involved in phonation - the tongue, respiratory muscles, facial muscles, pelvic floor, chest muscles, neck, back; exercises for relaxation of the larynx, massage of the vocal folds, meditation, exercises for physiological, light emission of sound, articulatory gymnastics, sound production at the marginal closure of the vocal folds, relaxation after prolonged vocal load, self-massage and stretching, vibration massage, tuning of resonators.

Below, as an example, three complexes for rehabilitation are given for three different pathologies: functional dysphonia of the hypotonic type, hypertonic voice disorders, vasomotor monochorditis.

Functional dysphonia of the hypotonic type.

1. Strelnikova's gymnastics 20 minutes daily

2. Active dance programs (aerobics, ZUMBA). A non-standard item, but often patients with this pathology have a decreased tone of the body as a whole, a sluggish emotional state. These fitness programs have a positive impact on the dynamics of rehabilitation. A well-proven method is also pronouncing short syllable rows for rhythmic training, for example, using the labial vibrant "BVI BVE BVA BVO BVU BVI".

3. Learning correct phonation on a soft attack

4. It is recommended to take a week daily for a week during breakfast - stir in 200 ml of water 180 mg of Vitamin C, Calcium 300 mg, 2 tablespoons of sugar, tincture of Eleutherococcus 20 drops.

5. A diet is set up - it is recommended to consume proteins at the beginning of the meal, then carbohydrates, which ensures a stable blood sugar level, and also helps to control weight, not to abuse fast carbohydrates, especially at the beginning of the meal.

6. Lymphatic drainage gymnastics.

7. Recommended phonopedic rehabilitation exercises - shtro-bass, inspirational phonation - to relax and massage the vocal folds, as well as to include the resonance of the tracheo-bronchial tree in the work.

8. Provide voice peace - reduced voice load in everyday life. Communicate with family members using a notebook and a pen. At work, ensure, if possible, the same conditions.

9. Full body stretching - 30 minutes a day.

10. Stretching the muscles of the neck and pectoralis major muscle with the simultaneous lowering of the larynx with the bitten lower lip on inhalation and exhalation on a silent "pfu". The phoneme "U" was chosen in order to avoid the larynx jump when changing inhalation with exhalation.

11. Voice training using the hardware-software complex is conducted six days a week, on the seventh day - rest.

Hypertensive voice disorders.

1. Jogging two or three times a week for at least half an hour, or brisk walking in the morning before the start of phonation to recycle excess adrenaline accumulated in the muscles.

2. Application of the following technique. We invite the patient to imagine that he is chewing, we ask him to perform chewing movements (without any food, water and chewing gum in the mouth, of course). In the process of chewing, reflex relaxation occurs, the person does not feel danger (adrenaline is a hormone produced in response to situations of danger), at the moment of chewing, we first begin to emit sounds on the phonation mechanism of moaning “Oh, how delicious!”, Then we introduce syllables, words, we increase the dynamic and intonation range of the voice.

3. Trainings to master the aspirated, then soft attack.

4. Stretching for the whole body - 30 minutes a day.

5. Stretching the muscles of the neck and pectoralis major muscle with the simultaneous lowering of the larynx with the bitten lower lip while inhaling and exhaling on a silent "pfu". The phoneme "U" was chosen in order to avoid the larynx jump when changing inhalation with exhalation.

Vasomotor monochorditis.

1. Inspiratory phonation

2. The administration of venotonics and angioprotectors - drugs detralex 500 mg or troxevasin in capsules is indicated.

3. Intra-laryngeal infusions are also shown: 1 ml. liquid aloe extract (solution for subcutaneous administration in ampoules 1 ml.), break 1 troxevasin capsule and pour out the powder, 3 ml. saline. After the infusion, it is recommended to be silent for 15 minutes, then phonopedic training is carried out.

4. Consultation of a gynecologist and an endocrinologist is recommended in order to normalize hormonal levels.

5. With this pathology, a month of silence is recommended (voice mode), then phonopedic training for bone-abdominal breathing begins, the correct technique of singing, and it is also possible to select a more suitable (often lower in tessiture) repertoire for the vocalist.

The method is carried out as follows. For this, four main blocks are sequentially performed, which are stages of the implementation of a technical solution:

1. Data collection unit

2. Data processing unit

3. Neural network training unit

4. Block of diagnostics and rehabilitation

The data collection unit is designed to collect sound samples of the voices of control patients and the corresponding pathologies of the vocal apparatus, as well as to record the voices of the reference sound of healthy people from the control group. Patient voice samples are recorded using a hardware-software complex. Voice samples are obtained by pronouncing different types of test items by the patient. Already at this stage, the test subjects (pronouncing) are divided into patients who use speech and singers. For each group, very specific tasks are set. When undergoing diagnostics, patients are divided into two types, depending on the scope of the voice function - a purely speech function or, if a person is a singer, then a vocal one. Accordingly, different types of tests are used to obtain voice samples for different types of patients. Voice samples and corresponding pathologies are stored in a database.

The data processing unit is designed to prepare input information for the neural network. For this, from the sound samples obtained at the first stage, information is extracted that is essential for determining the diagnosis and obtained from the analysis of frequency-amplitude characteristics, such as, for example, phonation duration, sound purity, strength, evenness, clarity, method of sound presentation. The hardware and software (in particular, Fourier analysis), used in the implementation of the method, allows us to measure, store and analyze all these characteristics.

The neural network training unit determines the mathematical correlation between the received sound data and the patient's diagnoses. Determining such a correlation requires a lot of computational work. The more preliminary data is collected, the more accurate the correlation can be obtained. The result of this block is a trained neural network.

Diagnostics and rehabilitation unit - the main unit, which is designed to diagnose patients and treat them. With the help of the hardware-software complex, the patient records samples of his voice. Then the neural network, based on the previously obtained samples, as well as using the established correlation, produces a diagnosis in the form of a list of probabilities of various pathologies. Based on this list and the results of a traditional clinical examination, the doctor makes a diagnosis. Thus, the accuracy of diagnosis of the disease is increased and the time of diagnosis is reduced. For the treatment of pathologies, the program provides a list of exercises and recommendations. This list was loaded into the program during the neural network training stage. Diagnosis of a patient using a hardware-software complex is carried out at a constant frequency (for example, once a day). Based on the sequence of diagnostic data, the dynamics of the patient's rehabilitation is determined. The doctor monitors the effectiveness of treatment based on the collected statistics and, if necessary, quickly corrects the rehabilitation process. The data collected by the system also allows clinicians to investigate the effectiveness of different treatments.

The claimed technical solution differs from analogs by the presence of a rehabilitation unit, which allows to optimize the restoration of the patient's voice using a neural network - to accelerate the recovery due to an objective daily assessment of the results of rehabilitation and timely correction, if necessary.

Example 1. Patient # 1, woman, 42 years old, bank employee.

Complaints - hoarseness, reduced endurance. After only five minutes of quiet conversation, the voice begins to wheeze.

Diagnostics was carried out using clinical commonly used methods: magnetic resonance imaging of the cervical spine, video stroboscopy of the larynx, narrow-spectrum imaging (NBI) and diagnostics using a hardware-software complex. Phonation recording was carried out in several stages, since the reduced endurance of the voice did not allow for diagnostics at one time.

Revealed - the range of speech voice is narrowed: from 195 Hz to 329 Hz. Weakness of the organs of articulation, tightness of the muscles of the neck, osteochondrosis of the cervical spine.

The diagnosis of the hardware-software complex is hypotonic dysphonia.

A consultation with a phoniatrist was held. The diagnosis was confirmed.

Along with the treatment usually used for this diagnosis, such as diet and drug treatment, prescribed by a phoniatrist (pp. 1-3, 5), rehabilitation measures recommended by the software and hardware complex and confirmed by a doctor were prescribed (pp. 4, 6 -7). The list of phonopedic exercises is recommended by the hardware-software complex.

The recommended complex for treatment and rehabilitation includes:

1. It is recommended to take a week daily during breakfast - stir in 200 ml of water 180 mg of Vitamin C, Calcium 300 mg, 2 tablespoons of sugar, tincture of Eleutherococcus 20 drops. Further, it was recommended to continue taking it every other day.

2. Set up a diet - it is recommended to consume proteins at the beginning of the meal, then carbohydrates, which ensures a stable blood sugar level, and also helps to control weight, not to overuse fast carbohydrates, especially at the beginning of the meal.

3. Conducted lymphatic drainage gymnastics, Strelnikova's gymnastics for 20 minutes in the morning.

4. Recommended phonopedic rehabilitation exercises - shtro-bass, inspiratory phonation - to relax and massage the vocal folds, as well as to include the resonance of the tracheo-bronchial tree in the work.

5. Provided voice peace - reduced voice load in everyday life. I communicated with family members using a notebook and a pen. At work, I tried to provide, if possible, the same conditions.

6. Stretching for the whole body - 30 minutes a day.

7. Stretching the muscles of the neck and pectoralis major muscle with the simultaneous lowering of the larynx with the bitten lower lip on inhalation and exhalation on a silent "pfu". The phoneme "U" was chosen in order to avoid the larynx jump when changing inhalation with exhalation.

The above appointments were carried out for 2.5 weeks.

Further, correction was made using PAK and the appointment of additional rehabilitation measures, taking into account individual characteristics and as the recovery process progressed. Voice training using PAK is carried out, starting from the third week of treatment, six days a week, the seventh day is rest.

Voice training using PAK consists in teaching the correct sound attack. Since hypotonia occurs if the coordination between the diaphragm and the larynx is disturbed, it is necessary to restore correct coordination by synchronizing the activation of the diaphragm and glottis when sound is attacked, which is achieved by first pronouncing short sounds: "MA, MO, MU, ME, MI", and then: "A, O, U, E, I" under the control of the hardware and software complex.

Control and success of execution were assessed using a software and hardware complex. The patient recorded her voice in a voice training teaching the correct sound attack using the PAK during daily practice. The software compared the patient's current voice recording with the voice recordings made at the initial diagnosis and with daily recordings, as well as with the recording of the reference exercise performance. In this case, the following characteristics of sound were used: fundamental frequency, energy distribution over the frequency spectrum, amplitude, rate of increase in amplitude, improving phonation, expanding the range, increasing voicing, improving timbre qualities, the presence of "flightiness" of the voice. The hardware and software complex also assesses the success of the rehabilitation tasks, issues a report to both the patient and sends it to the attending physician and phonopedist. The phonopedist and the attending physician adjust the recommendations of the PAK and the course of treatment at the beginning and during the rehabilitation process.

As a result of the treatment, including rehabilitation measures using PAA, an increase in voice endurance was achieved, the tone of the vocal cords was normalized, the correct motor stereotype of voice formation was established, which provides clarity of speech, "flight of the voice" and physiological work of the vocal apparatus.

The increase in voice endurance was assessed according to the patient's subjective feelings, and the duration of the voice training was measured before the onset of a feeling of slight voice fatigue. Sounds of vocal cords that are in the correct tone can be easily distinguished from variations with deviations. The flight of the voice is measured by the presence of a pronounced high singing formant in the spectrum of the voice, which lies in all - both male and female voices in the region of 2300-3500 Hz. Rehabilitation of this patient lasted one month. If without PAK, it would be much longer, since the standard period of hypotension rehabilitation is up to 4 months.

Example 2. The patient is a woman, 24 years old. Voice professional, vocalist. Soprano.

Complaints - stamina and voice range are reduced. The range of the voice before the onset of pathology in the small octave is 329 Hz - B-flat of the third 1864 Hz.

Diagnosed with special test exercises for singers.

Revealed - the range of the singing voice from mi of a small octave 329 Hz to mi of the second octave of 1318 Hz, the impossibility of phonation of the upper part of the range, the impossibility of modulating the voice in height, strength. The maximum phonation length is 8 sec. Falling out of minor third notes from "A" of the first to "C" of the second octave. The high singing formant is poorly expressed, the "flightiness" of the voice is reduced. Lethargy of the organs of articulation, stiffness of the muscles of the neck and shoulders, a general reduced energy potential.

Also, an emergency stressful situation, an incorrectly selected repertoire, hormonal disturbances, cervical osteochondrosis, a general high stressful environment, and a large voice load also indirectly influenced the development of pathology.

Diagnosis - vasomotor monochorditis, hypo-hypertonicity of the larynx. The diagnosis was confirmed by a phoniatrist.

Rehabilitation measures:

1. Reception daily Detralex 500 mg, vitamin C - 180 mg, coenzyme Q 60 mg, bromelain 500 mg with meals during the day.

2. Physiotherapy exercises 3 times a week, lymphatic drainage exercises, swimming in the pool.

3. Brisk walking every day 45 minutes before phonopedic workouts to recycle excess adrenaline in muscle tissue. Running for 30 minutes is possible. After running, there is a pause of 2-3 hours before singing.

4. Training regimen: 4 days of singing, 2 days of rest, then 5 days of singing, 2 days of rest, 5 days of singing, 1 day of rest, then, as you recover, 6 days of singing, 1 day of rest.

5. Rehabilitation exercises - shtro-bass, inspiratory phonation - to relax and massage the vocal folds, develop the marginal closure of the vocal folds.

6. Taking into account that the tone and speed of closure of the vocal folds in this patient are naturally high, the main task of rehabilitation is to develop a correct soft attack as opposed to a hard one.

7. Above dull tones in the middle section of the voice range, work on open vowels, trying to achieve an apparent false intensity and sonority, do not darken the sound. Ultimately, the vocal folds will get used to the correct mechanism and the sonority of the dropped tones will be restored.

8. Exercise in front of a mirror, removing unnecessary body movements and articulation apparatus while singing.

9. Shows general relaxation, meditations daily for 10 minutes.

10. Stretching the muscles of the neck and pectoralis major muscle.

11. In order to prevent voice overload: do not exercise too much, do not sing for more than two hours a day. An hour in the morning, then a long break and in the evening an additional 45-60 minutes.

Thus, the method provides a higher rate of rehabilitation and treatment success. Before the application of the hardware-software complex, patient # 2 could not cope with voice problems for a whole year due to the complexity of the pathology. Due to the instability of the vocal apparatus, the singer could not get a job as a soloist-vocalist. It is known that with a prolonged pathological state of the voice, the so-called "habitual hoarseness" occurs, i.e. the sound of an unhealthy apparatus, and even more so the imperfect method of voice formation, is firmly remembered by the vocalist, who is gradually giving up the hope of a cure. This complicates the rehabilitation process. With the use of the software and hardware complex, it became possible for the vocalist himself to visually control the correctness of the exercise due to the ability to track the formant pattern and the nature of the sound attack, which made it possible to correlate his sound representations with the objective data of the report of the software and hardware complex.

Thus, a new comprehensive method for the diagnosis and rehabilitation of patients with impaired voice-speech function using a trained neural network has been proposed, which expands the arsenal of known methods for the indicated purpose and allows not only to quickly and more accurately make a diagnosis, but also to restore the voice as quickly as possible and without relapse. speech function due to the implementation of specialized rehabilitation exercises under the continuous control of the hardware-software complex, which provides monitoring of the dynamics of rehabilitation and instant correction, if necessary.

Claims (2)

1. A method for the diagnosis and rehabilitation of patients with impaired voice-speech function, which includes collecting voice samples from control patients by pronouncing various types of test tasks by control patients, recording voice samples from control patients into a database using a software and hardware complex, determining frequency-amplitude voice characteristics by the Fourier analysis method, preparation of input information for a neural network, determination of mathematical correlation between the received sound data and patient diagnoses using a neural network, formation of a database of the dependence of the frequency-amplitude characteristics of the voice on voice-speech pathologies and the reference voice of control patients, implementation recordings using a software and hardware complex of samples of a patient's voice with impaired voice-speech function by pronouncing the same test tasks by him, analyzing the frequency-amplitude characteristics of the patient's voice, determining using software complex diagnosis of a patient by comparing the frequency-amplitude characteristics of the voice of a patient with impaired voice-speech function with the frequency-amplitude characteristics of the voice of control patients from the database and taking into account the dependence of the diagnosis of control patients on the frequency-amplitude characteristics of their voice and the implementation of rehabilitation, that in addition, the patient's rehabilitation procedures are carried out using complexes of exercises, including phonopedic gymnastics, selected by the hardware-software complex for specific pathologies, with constant monitoring of the dynamics of the rehabilitation process by objectively assessing the state of the patient's voice characteristics using the hardware-software complex by comparing the characteristics of the patient's voice with the characteristics his voice, obtained during the initial diagnosis and during the performance of rehabilitation exercises, as well as with the recording of the reference exercise performance and correction of the rehabilitation procedure depending on from the results of the dynamics of rehabilitation. 2. The method according to claim 1, characterized in that the database of voices of the hardware-software complex is formed for two types of patients with impaired voice-speech functions - for ordinary patients and for voice professionals - and contains two corresponding types of test exercise complexes for these types patients - to restore purely speech function or vocal function as well.

Images