A medical companion robot voice interaction recognition method and system

Abstract

The present application relates to the field of speech processing technology, and specifically to a speech interaction recognition method and system for a medical accompanying robot, which specifically includes: dividing the patient speech signal collected by the medical accompanying robot into frames to obtain each speech frame; constructing the medical noise confounding factor of the current speech frame based on the fluctuation characteristics of the signal strength in the current speech frame; constructing the medical noise matching index of the current speech frame based on the medical noise confounding factor of the speech frame of the current speech frame's neighbors and the peak characteristics of the signal strength in the corresponding time period; performing speech enhancement based on the medical noise matching index combined with a speech noise reduction algorithm; and interacting with the robot based on the enhanced speech of the patient. Thus, the speech interaction recognition of the medical accompanying robot is realized, the effect of enhancing the patient's speech signal is improved, the patient's speech part is clearer, and the patient's needs can be more accurately identified through the enhanced speech signal.

Description

Technical Field

[0001] The present application relates to the field of speech processing technology, and in particular to a speech interaction recognition method and system for a medical accompanying robot. Background Art

[0002] Medical companion robots are a new type of multifunctional service robots. They can provide services, safety monitoring, and human-computer interaction. They are mainly used to assist medical care workers, especially in caring for the elderly, the weak, or patients with special needs. In order to assist medical care workers in caring for patients more conveniently, accurately, and quickly, it is necessary to improve the voice interaction function in medical companion robots. Voice interaction mainly includes four key modules: voice recognition, information processing, command matching, and voice synthesis. Among them, the voice recognition module plays an important role in voice interaction. By converting the patient's voice into text data, it captures and understands the patient's needs, thereby achieving effective communication between man and machine.

[0003] In traditional speech enhancement algorithms, the principle of spectral subtraction to enhance the patient's speech signal is to use the power spectrum of the unvoiced sound as the power spectrum of the noise to achieve denoising and enhancement of the speech. However, the patient may be physically weak, so the patient's breath is relatively weak, the voice is relatively small, and the contrast with the noise is weak. In addition, the intensity of background noise is different in different environments in the hospital. For example, in the hospital lobby, the human voice is noisy and the background noise is strong. In the ward, the background noise is usually weak. Therefore, when using traditional spectral subtraction to enhance the patient's speech signal, different degrees of noise retention may occur, forming rhythmic and fluctuating music noise, resulting in the inability to accurately identify the relevant needs in the patient's voice, thereby affecting the execution and response effect of the medical accompanying robot. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of this application is to provide a medical companion robot voice interaction recognition method and system, and the technical solutions adopted are as follows:

[0005] In a first aspect, an embodiment of the present application provides a method for voice interaction recognition of a medical accompanying robot, the method comprising the following steps:

[0006] The medical accompanying robot collects the patient's voice signal and divides the voice signal into frames to obtain various voice frames;

[0007] Constructing a medical noise confounding factor of the current speech frame based on the fluctuation characteristics of the signal strength in the current speech frame;

[0008] Constructing a medical noise matching index of the current speech frame based on the medical noise confounding factor of the speech frame adjacent to the current speech frame and the peak characteristics of the signal strength in the corresponding time period;

[0009] Based on the medical noise matching index of the current speech frame, a medical noise suppression factor of the current speech frame is constructed, and speech enhancement is performed based on the medical noise suppression factor combined with a speech noise reduction algorithm;

[0010] Interact with the robot based on the patient's enhanced voice.

[0011] In one embodiment, the process of obtaining the medical noise confounding factor is:

[0012] Acquire a signal strength sequence of the current speech frame; calculate the mean of all elements in the signal strength sequence, recorded as the first mean; calculate the difference between each element in the signal strength sequence and the first mean, recorded as the first difference; and use the fusion value of the first difference of all elements in the signal strength sequence as the medical noise confounding factor of the current speech frame.

[0013] In one embodiment, the signal strength sequence is a sequence consisting of all signal strengths in the current speech frame.

[0014] In one embodiment, the process of obtaining the medical noise matching index is:

[0015] Taking a preset first number of speech frames closest to the current speech frame as adjacent speech frames of the current speech frame;

[0016] Constructing a confounding factor sequence of a current speech frame based on medical noise confounding factors of adjacent speech frames;

[0017] Constructing a neighboring peak time sequence of the current speech frame based on the peak characteristics of the signal strength in the neighboring speech frames;

[0018] Calculate the intonation weakness matching factor of the current speech frame based on the neighboring peak time sequence;

[0019] The fusion value of the intonation weakness matching factor and all elements in the confounding factor sequence is used as the medical noise matching index of the current speech frame.

[0020] In one embodiment, the confounding factor sequence is a sequence consisting of medical noise confounding factors of all adjacent speech frames of the current speech frame.

[0021] In one embodiment, the process of acquiring the neighboring peak time sequence is as follows:

[0022] The sequence consisting of all signal strengths in the time period of all adjacent speech frames of the current speech frame is taken as the first adjacent sequence; the first adjacent sequence is taken as the input of the peak search algorithm, and the sequence consisting of the time corresponding to all the output characteristic peaks is recorded as the neighboring peak time sequence.

[0023] In one embodiment, the process of obtaining the intonation weakness matching factor is as follows:

[0024] A first-order difference sequence of the neighboring peak time sequence is obtained, and the discreteness and average value of all elements in the first-order difference sequence are calculated, which are recorded as a first discreteness and a second average value, respectively; and a fusion value of the first discreteness and the second average value is used as the intonation weakness matching factor of the current speech frame.

[0025] In one embodiment, the process of obtaining the medical noise suppression factor is:

[0026] The fusion value of all signal intensities in the signal intensity sequence of the current speech frame and the medical noise matching index is used as the medical noise suppression factor of the current speech frame.

[0027] In one embodiment, the speech enhancement is performed based on the medical noise suppression factor combined with the speech noise reduction algorithm, specifically:

[0028] Each speech frame is used as the input of the speech noise reduction algorithm, wherein the medical noise suppression factor of each speech frame is used as the coefficient of the noise power spectrum, and the enhanced speech frames are output to obtain the enhanced speech signals.

[0029] In the second aspect, an embodiment of the present application also provides a medical accompanying robot voice interaction recognition system, including a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, the steps of any one of the above methods are implemented.

[0030] The embodiments of the present application have at least the following beneficial effects:

[0031] The present application constructs a medical noise confounding factor by analyzing the stationary characteristics of the speech signal when the patient speaks, reflecting the degree of confounding of the speech signal within the speech frame; based on the medical noise confounding factor, the sound intensity and interval characteristics of the patient when speaking are analyzed to construct a medical noise matching index, reflecting the degree to which the speech frame conforms to the noise distribution characteristics; based on the medical noise matching index, the noise intensity characteristics are analyzed to construct a medical noise suppression factor, reflecting the degree of need for noise suppression within the speech frame; the medical noise suppression factor is used as the coefficient of the noise power spectrum in the spectral subtraction method to achieve different degrees of suppression of noise in different environments, improve the effect of enhancing the patient's speech signal, make the patient's speech part clearer, and be able to more accurately identify the patient's needs through the enhanced speech signal, achieve effective communication between man and machine, enable the medical companion robot to execute corresponding instructions, and more accurately and timely assist medical care staff in taking care of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 A flowchart of a method for voice interaction recognition of a medical accompanying robot provided in one embodiment of the present application;

[0034] Figure 2 It is a flow chart of the voice interaction process;

[0035] Figure 3 Schematic diagram of the steps in the process of obtaining the medical noise matching index. DETAILED DESCRIPTION

[0036] In order to further explain the technical means and effects adopted by this application to achieve the predetermined invention purpose, the following is a detailed description of the medical accompanying robot voice interaction recognition method and system proposed in this application, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0037] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0038] The following is a detailed description of the specific scheme of the medical accompanying robot voice interaction recognition method and system provided by the present application in conjunction with the accompanying drawings.

[0039] See also Figure 1 , which shows a flowchart of a method for voice interaction recognition of a medical accompanying robot provided by an embodiment of the present application, the method comprising the following steps:

[0040] Step S1, collecting the patient's voice signal through a medical accompanying robot, and processing the voice signal into frames to obtain voice frames.

[0041] The medical accompanying robot's built-in microphone array collects the patient's voice signal in real time, and the sampling frequency can be set by the implementer. This application sets the sampling frequency to 8kHz.

[0042] Since the speech signal collected in the natural state contains various noise signals, the speech signal is first pre-emphasized. The pre-emphasis coefficient is taken as 0.96 in this application. It should be noted that the value of the pre-emphasis coefficient can be set by the implementer, and this embodiment does not make specific restrictions. Then the speech signal after pre-emphasis processing is input into the FIR high-pass filter, and the resolution of the high-frequency part of the speech signal is improved by the FIR high-pass filter, which effectively reduces the adverse effects of lip radiation. At the same time, since the movement of the vocal organ has motion inertia, it can be approximately considered that the speech signal is unchanged in a short time and has short-term stability. This characteristic is used and the Hamming window with a relatively complete spectrum preservation is selected for windowing and framing. The frame length is taken as 25ms, and the windowing and framing with a frame shift length of 15ms is performed. The frame after windowing and framing is recorded as a speech frame. It should be noted that the frame length and frame shift length can be set by the implementer, and this embodiment does not make specific restrictions. The speech signal after windowing and framing is used as the input of the double threshold method, and the speech signal with speech activity is output. Among them, FIR high-pass filter, Hamming window and double threshold method are well-known technologies, and the specific process is not repeated in this application. It should be noted that this application only provides a filtering method for filtering speech signals. There are many existing filtering methods. The implementer can also use other filtering algorithms to filter speech signals, and this application does not make specific restrictions.

[0043] Step S2, based on the fluctuation characteristics of the signal strength in the current speech frame, analyze the stability of the speech signal when the patient speaks, and construct the medical noise confounding factor of the current speech frame.

[0044] As an example, this step may include the following steps:

[0045] Since the movement of the vocal organs has inertia, and when the patient's body is relatively weak, there is a certain difference in the force of the throat compared to the healthy state, which causes changes in the patient's voice quality, including lower pitch, lower speaking rate, dragging sound and other phenomena. Therefore, after the patient's voice signal is processed by windowing and framing, the voice signal in each frame should be relatively stable. The noisy sounds in the hospital lobby include the voices of other people, the sound of shouting prompts, etc., so there may be a variety of sounds mixed together at the same time point, and the voice signal is relatively unstable. Based on the above analysis, this application constructs a medical noise confounding factor to reflect the degree of confounding of the voice signal in each frame. The construction process of the medical noise confounding factor can include the following steps:

[0046] (1) The sequence composed of the signal strengths of all sampling points in the current speech frame of the patient's speech signal is recorded as a signal strength sequence. Since the sampling frequency in this application is 8 kHz and the frame length is 25 ms, the length of the signal strength sequence is 200.

[0047] (2) Construct the medical noise confounding factor of the current speech frame based on the element fluctuation characteristics in the signal strength sequence of the current speech frame:

[0048] Calculate the mean of all elements in the signal strength sequence of the current speech frame, recorded as the first mean;

[0049] Calculate the difference between each element in the signal strength sequence of the current speech frame and the first mean value, and record it as the first difference;

[0050] The fused value of the first difference of all elements in the signal strength sequence of the current speech frame is used as the medical noise confounding factor of the current speech frame, wherein the medical noise confounding factor of the current speech frame is positively correlated with the fused value of the first difference.

[0051] Preferably, as an embodiment of the present application, the first difference may be the absolute value of the difference between each element in the signal strength sequence of the current speech frame and the first mean; the medical noise confounding factor of the current speech frame may be the cumulative sum of the first differences of all elements in the signal strength sequence of the current speech frame;

[0052] In other embodiments of the present application, the first difference may be the square of the difference between each element in the signal strength sequence of the current speech frame and the first mean; the medical noise confounding factor of the current speech frame may be the average value of the first differences of all elements in the signal strength sequence of the current speech frame.

[0053] In the signal strength sequence of the current speech frame, the more obvious the fluctuation of the signal strength is, the more likely it is that there are multiple sounds mixed together, and the more it conforms to the distribution characteristics of noise, so the calculated medical noise confounding factor is larger.

[0054] Step S3, constructing a medical noise matching index of the current speech frame based on the medical noise confounding factors of each speech frame adjacent to the current speech frame and the peak characteristics of the signal strength in the corresponding time period.

[0055] As an example, this step may include the following steps:

[0056] When a patient speaks, due to physical weakness, he may speak slower and louder, which is reflected in the voice signal as a lower overall signal strength. At the same time, the patient's breath is relatively weak, and he speaks slowly and pauses for a long time. In the voice signal, there is a certain distance between the characteristic peaks corresponding to the pronunciation, and the characteristic peak distances between different characteristic peaks are close. However, environmental noise is irregular and does not have this characteristic.

[0057] (1) Construct a confounding factor sequence of the current speech frame based on the medical noise confounding factors of the adjacent speech frames of the current speech frame:

[0058] First, all speech frames are arranged in ascending order according to the corresponding time sequence; then, for the current speech frame, a preset first number of speech frames with the closest Euclidean distance to the current speech frame are obtained as the adjacent speech frames of the current speech frame; finally, the sequence composed of the medical noise confounding factors of all adjacent speech frames of the current speech frame is recorded as the confounding factor sequence. It should be noted that the preset first number can be set by the implementer, and in this embodiment, the preset first number is set to 10.

[0059] (2) Based on the peak characteristics of the signal strength in the adjacent speech frames of the current speech frame, the neighboring peak time sequence of the current speech frame is constructed:

[0060] The sequence composed of all signal intensities in the time period of all neighboring speech frames of the current speech frame is taken as the first neighboring sequence; the first neighboring sequence is taken as the input of the automatic multi-scale peak search algorithm, and the output is the time corresponding to the characteristic peak in the first neighboring sequence, and the sequence composed of the time corresponding to all characteristic peaks is recorded as the neighboring peak time sequence. Among them, the automatic multi-scale peak search algorithm is a well-known technology, and the specific process is not repeated in this application. It should be noted that this application only provides a peak search method for peak search in the first neighboring sequence. There are many existing peak search methods. Implementers can also use other peak search algorithms to search for peaks in the first neighboring sequence, and this application does not make specific restrictions.

[0061] (3) Calculate the intonation weakness matching factor of the current speech frame based on the neighboring peak time sequence of the current speech frame:

[0062] A first-order difference sequence of the neighboring peak time sequence of the current speech frame is obtained, and the discreteness and average value of all elements in the first-order difference sequence are calculated, which are respectively recorded as the first discreteness and the second mean; a fusion value of the first discreteness and the second mean of the first-order difference sequence is used as the weak intonation matching factor of the current speech frame, wherein the weak intonation matching factor of the current speech frame is positively correlated with the second mean, and negatively correlated with the first discreteness.

[0063] Preferably, as an embodiment of the present application, the first discreteness may be the standard deviation of all elements in the first-order difference sequence; and the intonation weakness matching factor of the current speech frame may be the ratio of the second mean to the first discreteness.

[0064] In the neighboring peak time sequence of the current speech frame, the greater the difference in time intervals between different feature peaks, and the shorter the time interval, the less it conforms to the speech signal characteristics of the patient when speaking, so the calculated intonation weakness matching factor is smaller.

[0065] (4) Calculate the medical noise matching index of the current speech frame based on the intonation weakness matching factor and the confounding factor sequence of the current speech frame:

[0066] The fusion value of the intonation weakness matching factor of the current speech frame and all medical noise confounding factors in the confounding factor sequence is taken as the medical noise matching index of the current speech frame, wherein the medical noise matching index of the current speech frame is positively correlated with each medical noise confounding factor in the confounding factor sequence of the current speech frame, and negatively correlated with the intonation weakness matching factor of the current speech frame.

[0067] Preferably, as an embodiment of the present application, the average value of all medical noise confounding factors in the confounding factor sequence of the current speech frame is calculated; the medical noise matching index of the current speech frame may be the ratio of the average value of the medical noise confounding factors to the intonation weakness matching factor of the current speech frame;

[0068] In other embodiments of the present application, the calculation result of an exponential function with a natural constant as the base and the negative number of the intonation weakness matching factor of the current speech frame as the exponent is calculated; the medical noise matching index of the current speech frame can be the product of the average value of the medical noise confounding factor and the calculation result.

[0069] In the confounding factor sequence of the current speech frame, the larger the medical noise confounding factor of different speech frames, the more the signal strength change in the adjacent speech frames of the current speech frame conforms to the distribution law of noise, and the less likely the time period is when the patient is speaking; at the same time, the smaller the weak intonation matching factor is, the less it conforms to the speech signal characteristics when the patient is speaking, so the calculated medical noise matching index is larger. The medical noise matching index reflects the degree to which the current speech frame conforms to the noise characteristics.

[0070] Step S4, constructing a medical noise suppression factor of the current speech frame based on the medical noise matching index of the current speech frame, and performing speech enhancement.

[0071] As an example, this step may include the following steps:

[0072] (1) The medical noise matching index of each speech frame in the patient's speech signal obtained through the above steps reflects the degree to which each speech frame conforms to the noise distribution characteristics. In order to achieve a better enhancement effect on the patient's speech signal and avoid the appearance of music noise after enhancement, it is necessary to analyze the noise intensity of different speech frames and suppress noise of different intensities to different degrees. Based on the above analysis, this application constructs a medical noise suppression factor based on the medical noise matching index to reflect the degree of demand for noise suppression in each speech frame:

[0073] Obtaining each signal strength in the signal strength sequence of the current voice frame;

[0074] The fusion value of all signal intensities in the signal intensity sequence of the current speech frame and the medical noise matching index is used as the medical noise suppression factor of the current speech frame, wherein the medical noise suppression factor of the current speech frame is positively correlated with all signal intensities in the signal intensity sequence of the current speech frame and the medical noise matching index, respectively.

[0075] Preferably, as an embodiment of the present application, the medical noise suppression factor of the current speech frame may be the product of the mean of all signal strengths in the signal strength sequence of the current speech frame and the medical noise matching index;

[0076] In other embodiments of the present application, the medical noise suppression factor of the current speech frame may be the sum of the normalized value of all signal strength mean values ​​in the signal strength sequence of the current speech frame and the normalized value of the medical noise matching index.

[0077] The larger the medical noise matching index of the current speech frame, the more the current speech frame conforms to the characteristics of the noise distribution in the hospital. At the same time, the stronger the noise intensity in the signal intensity sequence, the greater the degree of suppression of the current speech frame should be, enhancing the contrast between the speech signal and the noise when the patient speaks, so the calculated medical noise suppression factor is larger.

[0078] (2) Each speech frame in the patient's speech signal is used as the input of the spectral subtraction method, and the medical noise suppression factor of each speech frame is used as the coefficient of the noise power spectrum. The noise of different speech frames is adjusted and the output is the enhanced speech frames, that is, the enhanced speech signal.

[0079] Step S5, interacting with the robot based on the patient's enhanced voice.

[0080] After obtaining the enhanced voice signal of the patient through the above steps, the subsequent steps can be carried out to complete the interaction between the patient and the robot. The voice interaction process flow chart is as follows: Figure 2 shown.

[0081] The specific steps for patients to interact with robot voice are as follows:

[0082] Speech recognition module: The enhanced speech signal of the patient is obtained through the above steps, and the enhanced speech signal is used as the input of the hidden Markov model, and the output is the text information corresponding to the patient's speech signal.

[0083] Information processing module: The text information output by the speech recognition module is used as the input of the BERT model, and the intent category in the text information is output for the medical accompanying robot to understand, that is, the text expression of the patient's needs.

[0084] Instruction matching module: According to the intention categories output in the information processing module, they are matched with different functions of the medical accompanying robot to enable the robot to complete the corresponding instructions.

[0085] Speech synthesis module: The intent category output in the information processing module is used as the input of the GPT model, and a text answer matching the patient is output. The generated text answer is used as the input of the WaveNet deep learning model, and the speech corresponding to the text answer is output and played through the built-in speaker of the medical companion robot to answer the patient's questions or needs.

[0086] In each module in the voice interaction process, the hidden Markov model, BERT model, GPT model and WaveNet deep learning model are well-known technologies, and the specific process will not be repeated in this application.

[0087] The schematic diagram of the steps in obtaining the medical noise matching index is as follows: Figure 3 shown.

[0088] Based on the same inventive concept as the above method, an embodiment of the present application also provides a medical escort robot voice interaction recognition system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned medical escort robot voice interaction recognition methods are implemented.

[0089] In summary, the embodiment of the present application provides a method for voice interaction recognition of a medical accompanying robot, which constructs a medical noise confounding factor by analyzing the stationary characteristics of the voice signal when the patient speaks, reflecting the degree of confounding of the voice signal in the voice frame; based on the medical noise confounding factor, the sound intensity and interval characteristics when the patient speaks are analyzed to construct a medical noise matching index, reflecting the degree to which the voice frame conforms to the noise distribution characteristics; based on the medical noise matching index, the noise intensity characteristics are analyzed to construct a medical noise suppression factor, reflecting the degree of need for noise suppression in the voice frame; the medical noise suppression factor is used as the coefficient of the noise power spectrum in the spectral subtraction method to achieve different degrees of suppression of noise in different environments, improve the effect of enhancing the patient's voice signal, make the patient's voice part clearer, and be able to more accurately identify the patient's needs through the enhanced voice signal, achieve effective communication between man and machine, enable the medical accompanying robot to execute corresponding instructions, and more accurately and timely assist medical caregivers in taking care of patients.

[0090] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. The above-mentioned specific embodiments of the present application are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0091] The various embodiments in the present application are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0092] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present application should be included in the protection scope of the present application.

Claims

1. A medical accompanying robot voice interaction recognition method, characterized in that: The method comprises the following steps: The medical accompanying robot collects the patient's voice signal and divides the voice signal into frames to obtain various voice frames; Constructing a medical noise confounding factor of the current speech frame based on the fluctuation characteristics of the signal strength in the current speech frame; Constructing a medical noise matching index of the current speech frame based on the medical noise confounding factor of the speech frame adjacent to the current speech frame and the peak characteristics of the signal strength in the corresponding time period; Based on the medical noise matching index of the current speech frame, a medical noise suppression factor of the current speech frame is constructed, and speech enhancement is performed based on the medical noise suppression factor combined with a speech noise reduction algorithm; Interact with the robot based on the patient’s enhanced voice; The process of obtaining the medical noise matching index is as follows: Taking a preset first number of speech frames closest to the current speech frame as adjacent speech frames of the current speech frame; Constructing a confounding factor sequence of a current speech frame based on medical noise confounding factors of adjacent speech frames; Constructing a neighboring peak time sequence of the current speech frame based on the peak characteristics of the signal strength in the neighboring speech frames; Calculate the intonation weakness matching factor of the current speech frame based on the neighboring peak time sequence; The fusion value of all elements in the confounding factor sequence and the intonation weakness matching factor is used as the medical noise matching index of the current speech frame.

2. A medical accompanying robot voice interaction recognition method as claimed in claim 1, characterized in that: The process of obtaining the medical noise confounding factor of the current speech frame is as follows: Acquire a signal strength sequence of the current speech frame; calculate the mean of all elements in the signal strength sequence, recorded as the first mean; calculate the difference between each element in the signal strength sequence and the first mean, recorded as the first difference; and use the fusion value of the first difference of all elements in the signal strength sequence as the medical noise confounding factor of the current speech frame.

3. A medical accompanying robot voice interaction recognition method as claimed in claim 2, characterized in that: The signal strength sequence is: a sequence composed of all signal strengths in the current speech frame.

4. A medical accompanying robot voice interaction recognition method as claimed in claim 1, characterized in that: The confounding factor sequence is a sequence consisting of medical noise confounding factors of all adjacent speech frames of the current speech frame.

5. A medical accompanying robot voice interaction recognition method as claimed in claim 1, characterized in that: The process of obtaining the neighbor peak time sequence is as follows: The sequence consisting of all signal strengths in the time period of all adjacent speech frames of the current speech frame is taken as the first adjacent sequence; the first adjacent sequence is taken as the input of the peak search algorithm, and the sequence consisting of the time corresponding to all the output characteristic peaks is recorded as the neighboring peak time sequence.

6. A medical accompanying robot voice interaction recognition method as claimed in claim 1, characterized in that: The process of obtaining the intonation weakness matching factor is as follows: A first-order difference sequence of the neighboring peak time sequence is obtained, and the discreteness and average value of all elements in the first-order difference sequence are calculated, which are recorded as a first discreteness and a second average value, respectively; and a fusion value of the first discreteness and the second average value is used as the intonation weakness matching factor of the current speech frame.

7. A medical accompanying robot voice interaction recognition method as claimed in claim 2, characterized in that: The process of obtaining the medical noise suppression factor is as follows: The fusion value of all signal intensities in the signal intensity sequence of the current speech frame and the medical noise matching index is used as the medical noise suppression factor of the current speech frame.

8. A medical accompanying robot voice interaction recognition method as claimed in claim 1, characterized in that: The speech enhancement based on the medical noise suppression factor combined with the speech noise reduction algorithm is specifically: Each speech frame is used as the input of the speech noise reduction algorithm, wherein the medical noise suppression factor of each speech frame is used as the coefficient of the noise power spectrum, and the enhanced speech frames are output to obtain the enhanced speech signals.

9. A medical accompanying robot voice interaction recognition system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images