A system and method for synthesizing speech for the elderly based on gerontology

Abstract

The application discloses a kind of based on old-age care old person synthetic voice system, method, based on old-age care old person synthetic voice system, the old fuzzy voice acquisition and pretreatment module, for collecting old person pronunciation fuzzy, pronunciation unclear, sentence discontinuous original voice signal, and original voice signal is carried out noise reduction, volume gain, discontinuous segment splicing pretreatment;The old fuzzy voice recognition and semantic carding module is connected with old fuzzy voice acquisition and pretreatment module, for the acoustic feature extraction of preprocessed voice signal, old-age care exclusive semantic matching.This application solves both the problem that old people speak fuzzy and others cannot understand, and the problem that old people have poor hearing and cannot understand regular voice, bidirectional adaptation, covers all old-age care communication scene.Recognition and synthesis are no longer independent, the more fuzzy the pronunciation of the old people, the lower the average semantic weight, the higher the compensation level of synthetic voice, automatically slow down the speed, lengthen the pause, and strengthen the low frequency.

Description

Technical Field

[0001] This invention relates to the fields of speech signal processing, intelligent human-computer interaction, and elderly care assistive devices. Specifically, it relates to a system and method for synthesizing speech for the elderly based on elderly care, applicable to home-based elderly care, institutional elderly care, and hospital care scenarios. It specifically addresses the two-way communication barriers faced by the elderly due to muscle degeneration, neurological disorders, unclear pronunciation, and fragmented speech that are difficult for others to understand, as well as hearing loss that makes it difficult for the elderly to understand regular-speed speech. Background Technology

[0002] As the aging population continues to deepen, the demand for health assistance and intelligent interaction among the elderly is growing rapidly. Currently, most voice technologies on the market and in published patents for elderly care are one-way output modes, that is, they are designed only with slow, low-pitched synthetic voice reminders for medication reminders, daily routine announcements, and health education. The purpose is only to make the language "understandable" to the elderly, completely ignoring the core pain point that the elderly "cannot speak clearly".

[0003] Many elderly people, including those with sequelae of cerebral infarction, Parkinson's disease, or oral degeneration, commonly exhibit problems such as slurred speech, missing syllables, fragmented sentences, and disjointed expression. Family members, caregivers, and medical staff often find it difficult to accurately understand the elderly person's needs, which can easily lead to delays in addressing their concerns, emotional anxiety, and safety hazards.

[0004] Meanwhile, traditional speech recognition and speech synthesis operate independently, with no data association or parameter linkage: the recognition module is only responsible for transcribing text, and the synthesis module has a fixed set of elderly speech speed and tone, and will not dynamically adjust the broadcast strategy according to the clarity of the elderly’s own pronunciation; there is no cross-module quantitative linkage mechanism, and there is no adaptive logic that “the more unclear the elderly’s speech is, the more considerate and stronger the system’s feedback speech becomes.”

[0005] In addition, traditional technologies mostly rely on qualitative adjustments without standardized mathematical model constraints, resulting in unstable recognition accuracy and inconsistent speech adaptation effects, making them unsuitable for large-scale deployment in elderly care scenarios. Therefore, there is an urgent need to design a two-way closed-loop, modularly linked, and dual-formula quantification constraint-based synthetic speech system and method for the elderly. Summary of the Invention

[0006] To address the shortcomings of existing technologies, such as one-way interaction, fragmented modules, lack of quantitative control, weak fuzzy speech processing capabilities, and lack of adaptive compensation, this invention provides a system and method for synthesizing speech for the elderly based on elderly care. It establishes a five-module closed-loop architecture and achieves fuzzy speech semantic quantification correction and speech prosody adaptive reverse compensation through two interrelated mathematical formulas. It generates new and non-obvious technical effects through module linkage, while taking into account the dual needs of fuzzy speech translation for the elderly and age-appropriate adaptation of external speech.

[0007] This invention is implemented as follows: a system for synthesizing speech for the elderly based on elderly care, comprising: an elderly fuzzy speech acquisition and preprocessing module, an elderly fuzzy speech recognition and semantic sorting module, a semantic weight dynamic association module, an elderly-adaptive speech synthesis module, and a terminal interaction module; the elderly fuzzy speech acquisition and preprocessing module is used to acquire raw speech signals of elderly people with fuzzy pronunciation, unclear articulation, and discontinuous sentences, and to perform noise reduction, volume gain, and discontinuous segment splicing preprocessing on the raw speech signals; the elderly fuzzy speech recognition and semantic sorting module is communicatively connected to the elderly fuzzy speech acquisition and preprocessing module, and is used to extract acoustic features from the preprocessed speech signals, perform elderly-specific semantic matching, correct the fragmented and pronunciation-deviation-incorrect speech expressions of the elderly, and output standardized and accurate semantic text and semantic weight coefficients corresponding to each speech segment; the semantic weight dynamic association module... The system is electrically connected to the elderly-friendly fuzzy speech recognition and semantic sorting module and the elderly-adaptive speech synthesis module, respectively. It receives the semantic weight coefficients of all speech segments, calculates and generates an average semantic weight coefficient, and transmits this average semantic weight coefficient to the elderly-adaptive speech synthesis module in real time, achieving quantitative linkage between the recognition and synthesis sides. The elderly-adaptive speech synthesis module receives the average semantic weight coefficient from the semantic weight dynamic association module and adaptively adjusts the speech prosody parameters based on the average semantic weight coefficient, converting standard text or standard speech input from the outside into customized synthesized speech adapted to the hearing degeneration characteristics of the elderly. The terminal interaction module establishes data interaction with each of the aforementioned modules, enabling elderly speech input, local playback of elderly-adaptive synthesized speech, standardized semantic text visualization, and bidirectional information transmission and interaction between multiple terminals.

[0008] As a preferred embodiment of the present invention, the elderly fuzzy speech acquisition and preprocessing module has a built-in adaptive noise reduction unit specifically designed for elderly care scenarios. In response to the characteristics of elderly people speaking at low volumes, having a lot of oral airflow noise, and having a noisy background environment in home / elderly care institutions, it selectively filters out invalid environmental noise and human voice noise, thereby improving the signal-to-noise ratio of the acquired speech signal to more than 25dB. At the same time, it fully preserves the core low-frequency acoustic features of the elderly's voices, avoiding the loss of effective fuzzy speech features.

[0009] As a preferred embodiment of the present invention, the elderly fuzzy speech recognition and semantic sorting module has a built-in elderly-specific fuzzy speech dataset and semantic correction model. The semantic correction model is configured with a semantic correction formula for elderly fuzzy speech, and completes the semantic correction and integration of fragmented fuzzy speech through quantitative calculation. The semantic correction formula for ambiguous speech in the elderly is as follows: ;in, The overall score is based on semantic correction. The total number of independent speech segments for decomposing ambiguous speech in the elderly. For the first The semantic weight coefficients of each speech segment. For the original acoustic feature values ​​of fuzzy speech, Matching score for a semantic database specifically designed for elderly care; These are the speech feature weighting coefficients, with a value range of... This refers to the semantic matching weight coefficient, with a value range of... This is the noise interference penalty coefficient, with a value range of... This is the quantized value of the noise features corresponding to a single speech segment.

[0010] As a preferred embodiment of the present invention, the semantic weight coefficient The system comprehensively quantifies and judges based on the pronunciation completeness, articulation clarity, and sentence coherence of a single speech segment. The range of values ​​is The more unclear the pronunciation and the more severe the speech breaks of an elderly person, the higher the semantic weight coefficient of the corresponding segment. The smaller the value, the clearer and more complete the elderly person's pronunciation and expression, and the higher the semantic weight coefficient of the corresponding segment. The larger the value.

[0011] As a preferred embodiment of the present invention, the semantic weight dynamic association module retrieves the semantic weight coefficients of all speech segments. The average semantic weight coefficient is calculated using the arithmetic mean algorithm. ; ; The average semantic weight coefficient As a core quantization parameter for cross-module association, it is sent unidirectionally and in real time to the elderly-adaptive speech synthesis module, realizing closed-loop linkage from semantic recognition results to the speech synthesis control end.

[0012] As a preferred embodiment of the present invention, the elderly-adaptive speech synthesis module incorporates an elderly-adaptive speech prosodic adjustment model. This model is configured with an elderly-adaptive speech prosodic adjustment formula, which, combined with the received average semantic weight coefficients, performs adaptive quantization adjustment of speech rate, pauses, and pitch. The elderly-adaptive speech prosodic adjustment formula is as follows: in, This is the comprehensive prosody regulation compensation value. The average semantic weight coefficient output by the semantic weight dynamic association module; Fixed coefficients are used to fit hearing to older adults, with a range of values. This is a benchmark value for standard speaking speed; This is a statement length adjustment coefficient, with a value range of... The length of the statement in the text to be synthesized.

[0013] As a preferred embodiment of the present invention, the prosodic comprehensive regulation compensation value With average semantic weight coefficient A negative correlation mapping relationship exists; average semantic weight coefficient The smaller the value, the better the overall rhythm regulation compensation value. The larger the average semantic weight coefficient, the faster the synthesized speech rate, the longer the pause intervals, the stronger the low-frequency human voice gain, and the weaker the high-frequency sharp timbre; The larger the value, the greater the overall rhythm regulation compensation value. The smaller the volume, the more the synthesized speech will restore the normal speaking speed and natural rhythm, avoiding sluggish broadcasting.

[0014] As a preferred embodiment of the present invention, a cross-module adaptive compensation mechanism is formed through the coordinated linkage of multiple modules such as acquisition, recognition, weight association, and speech synthesis. The more unclear the elderly person's own pronunciation and the more obvious the expression impairment, the higher the auditory compensation level of the system for the feedback synthesized speech. While improving the accuracy of fuzzy speech recognition, it also improves the elderly person's ability to understand the broadcast speech.

[0015] As a preferred embodiment of the present invention, the terminal interaction module is divided into a local interaction terminal for the elderly and a remote interaction terminal for caregivers. The local interaction terminal for the elderly is equipped with large buttons, a high-contrast display interface, and extra-large fonts, and supports one-click voice input and extremely simple operation. The remote interaction terminal for caregivers can be any of a mobile phone, tablet, or computer, and can receive standard semantic text converted from the elderly's fuzzy speech in real time. It supports remote text / voice replies from family members, caregivers, and medical staff, realizing two-way barrier-free interaction in the elderly care scenario.

[0016] A method for implementing synthesized speech for the elderly based on elderly care, applied to the aforementioned system for synthesized speech for the elderly based on elderly care, includes the following steps: Step S1: An elderly fuzzy speech acquisition and preprocessing module acquires the original speech signal of the elderly person in real time, which is characterized by unclear pronunciation and fragmented speech, performs noise reduction, volume gain, and sentence segment splicing preprocessing, and outputs an optimized and clean speech signal; Step S2: An elderly fuzzy speech recognition and semantic sorting module extracts acoustic features from the optimized speech and matches it with an elderly care semantic database, substitutes it into a semantic correction formula to complete quantitative correction, outputs standard and fluent semantic text, and simultaneously generates semantic weight coefficients corresponding to each speech segment; Step S3: Semantic weights... The dynamic association module collects all semantic weight coefficients, calculates the average semantic weight coefficient, and encrypts and pushes the association parameter to the elderly-adaptive speech synthesis module in real time; Step S4: The elderly-adaptive speech synthesis module retrieves the average semantic weight coefficient, substitutes it into the prosody control formula to calculate the prosody comprehensive control compensation value, and adaptively adjusts the speech rate, pauses, and low-frequency gain parameters according to the compensation value to generate customized synthesized speech adapted to the auditory characteristics of the elderly; Step S5: The terminal interaction module pushes the standard semantic text to the caregiver's remote interaction terminal, and at the same time completes the gentle and adapted speech broadcast on the elderly's local interaction terminal, repeating the process to continuously complete the two-way closed-loop voice interaction between the elderly and the caregiver.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. It solves both the problem of elderly people's unclear speech being incomprehensible and the problem of elderly people with poor hearing being unable to understand regular speech, with bidirectional adaptation, covering all communication scenarios for the elderly. Recognition and synthesis are no longer independent. The more unclear the elderly person's pronunciation, the lower the average semantic weight, and the higher the level of synthesized speech compensation, automatically slowing down the speech speed, lengthening pauses, and amplifying low frequencies.

[0018] 2. Abandoning traditional qualitative adjustments, it relies on mathematical models to accurately calculate semantic scores and prosodic compensation values, ensuring stable operation, reproducibility, and ease of product deployment and parameter iteration. The recognition logic is optimized for fragmented speech, incomplete pronunciation, and high-frequency elderly care needs, automatically sorting and integrating fragmented sentences, significantly improving transcription accuracy. The user interface is extremely simple for seniors, while caregivers can view text in real time, adapting to different user abilities and demonstrating strong industrial applicability. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure provided in an embodiment of the present invention; Figure 2 This is provided by an embodiment of the present invention; Detailed Implementation

[0020] To further understand the invention's content, features, and effects, the following embodiments are provided, and detailed descriptions are given in conjunction with the accompanying drawings.

[0021] The structure of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] like Figures 1 to 2 As shown in the figure, an embodiment of the present invention provides a system for synthesizing speech for the elderly based on elderly care, characterized by comprising: an elderly fuzzy speech acquisition and preprocessing module, an elderly fuzzy speech recognition and semantic sorting module, a semantic weight dynamic association module, an elderly-adapted speech synthesis module, and a terminal interaction module; the elderly fuzzy speech acquisition and preprocessing module is used to acquire the original speech signal of the elderly with fuzzy pronunciation, unclear articulation, and discontinuous sentences, and to perform noise reduction, volume gain, and discontinuous segment splicing preprocessing on the original speech signal; the elderly fuzzy speech recognition and semantic sorting module is communicatively connected to the elderly fuzzy speech acquisition and preprocessing module, and is used to extract acoustic features and perform elderly-specific semantic matching on the preprocessed speech signal, correct the fragmented and pronunciation deviation speech expression of the elderly, and output standardized and accurate semantic text and semantic weight coefficients corresponding to each speech segment; the semantic weight dynamic association module... The association module is electrically connected to the elderly-friendly fuzzy speech recognition and semantic sorting module and the elderly-adaptive speech synthesis module, respectively. It is used to receive the semantic weight coefficients of all speech segments, calculate and generate the average semantic weight coefficient, and transmit the average semantic weight coefficient to the elderly-adaptive speech synthesis module in real time, realizing the quantitative linkage between the two major functional modules of recognition and synthesis. The elderly-adaptive speech synthesis module is coupled to receive the average semantic weight coefficient issued by the semantic weight dynamic association module. It is used to adaptively adjust the speech prosody parameters based on the average semantic weight coefficient, and convert the standard text or standard speech input from the outside into customized synthesized speech adapted to the hearing degeneration characteristics of the elderly. The terminal interaction module establishes data interaction with each of the above modules, and is used to realize the elderly's speech input, local broadcast of elderly-adaptive synthesized speech, standardized semantic text visualization display, and two-way information transmission and interaction between multiple terminals.

[0023] This system adopts a closed-loop linkage architecture. The acquisition module serves as the front-end entry point, specifically adapted for the low volume and noise-laden, degraded speech of the elderly. The recognition module is the core processing hub, completing the semantic repair of ambiguous speech. The semantic weight dynamic association module is the innovative core hub of this invention, breaking the traditional design concept of isolating recognition and synthesis, and realizing cross-module parameter flow. The elderly-adapted speech synthesis module is the execution output end, dynamically changing the broadcast style according to the upstream quantization parameters. The terminal interaction module carries the human-computer interface and bidirectional data distribution. The five modules cooperate with each other and communicate with each other to jointly support the bidirectional business of converting ambiguous speech to text and standard speech to elderly-adapted speech.

[0024] The elderly fuzzy speech acquisition and preprocessing module incorporates a dedicated adaptive noise reduction unit for elderly care scenarios. Addressing the characteristics of elderly people's low speaking volume, numerous oral airflow noises, and complex background interference in home / elderly care facility environments, it selectively filters out invalid environmental noise and human voice clutter, improving the signal-to-noise ratio of the acquired speech signal to over 25dB. Simultaneously, it fully preserves the core low-frequency acoustic features of the elderly's voice, avoiding the loss of effective fuzzy speech features. This limits the noise reduction and enhancement capabilities of the acquisition and preprocessing module. Elderly care home environments often contain television noise, wind noise, footsteps, and the sound of tableware clattering. Additionally, elderly people speak with heavy breathing and numerous oral airflow noises. This module employs an adaptive noise reduction algorithm to specifically suppress environmental noise and non-human voice clutter, preserving the low-frequency human voice features of the elderly, and stably improving the signal-to-noise ratio to over 25dB. This avoids excessive noise reduction that could cause the loss of incomplete syllables, providing a high-quality raw signal foundation for subsequent fuzzy speech recognition.

[0025] The elderly fuzzy speech recognition and semantic sorting module has a built-in fuzzy speech dataset and semantic correction model specifically for the elderly. The semantic correction model is configured with a semantic correction formula for elderly fuzzy speech, and completes the semantic correction and integration of fragmented fuzzy speech through quantitative calculation. The semantic correction formula for ambiguous speech in the elderly is as follows: (1) Among them, The overall score is based on semantic correction. The total number of independent speech segments for decomposing ambiguous speech in the elderly. For the first The semantic weight coefficients of each speech segment. For the original acoustic feature values ​​of fuzzy speech, Matching score for a semantic database specifically designed for elderly care; These are the speech feature weighting coefficients, with a value range of... This refers to the semantic matching weight coefficient, with a value range of... This is the noise interference penalty coefficient, with a value range of... This is the quantized value of the noise features corresponding to a single speech segment.

[0026] The module incorporates a massive database of ambiguous speech samples from the elderly, covering various types of speech impairments, repetitive phrases, and fragmented expressions from seniors with chronic diseases, disabilities, and advanced age. The formula described above functions as follows: it scores each segment of speech based on acoustic features and a matching lexicon for elderly language, deducts noise interference scores, and accumulates these to obtain a comprehensive semantic score. A higher score indicates more accurate semantic judgment, enabling automatic word order correction, completion of incomplete semantics, and correction of grammatical errors.

[0027] The semantic weight coefficient The system comprehensively quantifies and judges based on the pronunciation completeness, articulation clarity, and sentence coherence of a single speech segment. The range of values ​​is The more unclear the pronunciation and the more severe the speech breaks of an elderly person, the higher the semantic weight coefficient of the corresponding segment. The smaller the value, the clearer and more complete the elderly person's pronunciation and expression, and the higher the semantic weight coefficient of the corresponding segment. The larger the value.

[0028] This claim defines the determination rules and value logic for the semantic weight coefficient ωi. ωi is a core variable for cross-module linkage, entirely determined by the elderly person's real-time speaking status: clear pronunciation and complete sentences indicate a higher weight (0.8–1.0); unclear pronunciation, severe phrasing, and missing pronunciation indicate a lower weight (0.1–0.3). This parameter objectively quantifies the degree of speech impairment in the elderly person, providing a basis for subsequent reverse compensation.

[0029] The semantic weight dynamic association module retrieves the semantic weight coefficients of all speech segments. The average semantic weight coefficient is calculated using the arithmetic mean algorithm. ; ; The average semantic weight coefficient As a core quantization parameter for cross-module association, it is sent unidirectionally and in real time to the elderly-adaptive speech synthesis module, realizing closed-loop linkage from semantic recognition results to the speech synthesis control end.

[0030] This invention adds an average weight calculation formula to achieve the integration of multiple speech data segments. Since the weights of individual segments fluctuate greatly and have weak reference value, the global average semantic weight is obtained through arithmetic averaging and uniformly input into the synthesis module. This ensures overall smooth prosodic adjustment without abrupt changes in single sentences, achieving global adaptive control and completing the data integration between Formula 1 and Formula 2.

[0031] Furthermore, the elderly-adaptive speech synthesis module incorporates an elderly-adaptive speech prosody control model. This model is configured with an elderly-adaptive speech prosody control formula, which, combined with the received average semantic weight coefficients, performs adaptive quantization adjustment of speech rate, pauses, and pitch. The elderly-adaptive speech prosody control formula is as follows: (2) in, This is the comprehensive prosody regulation compensation value. The average semantic weight coefficient output by the semantic weight dynamic association module; Fixed coefficients are used to fit hearing to older adults, with a range of values. This is a benchmark value for standard speaking speed; This is a statement length adjustment coefficient, with a value range of... The length of the statement in the text to be synthesized.

[0032] This formula is a reverse correlation design, which is strongly bound to Formula 1: using the upstream output ωavg as the core independent variable, combined with the listening adaptation coefficient, speech rate benchmark, and text length, the final prosodic compensation value P is calculated; P directly corresponds to the speech rate multiplier, pause duration, and low-frequency gain intensity, and is the only quantitative basis for adjusting the synthesized speech parameters.

[0033] Specifically, the prosodic comprehensive regulation compensation value With average semantic weight coefficient A negative correlation mapping relationship exists; average semantic weight coefficient The smaller the value, the better the overall rhythm regulation compensation value. The larger the average semantic weight coefficient, the faster the synthesized speech rate, the longer the pause intervals, the stronger the low-frequency human voice gain, and the weaker the high-frequency sharp timbre; The larger the value, the greater the overall rhythm regulation compensation value. The smaller the volume, the more the synthesized speech will restore the normal speaking speed and natural rhythm, avoiding sluggish broadcasting.

[0034] The smaller ωavg is, the larger P is, resulting in stronger compensation: ultra-slow speed, long pauses, and low frequencies, suitable for elderly people with severe speech impairments; the larger ωavg is, the smaller P is, resulting in weaker compensation: moderate speech speed and natural rhythm, avoiding listening fatigue and sluggish broadcasting for healthy elderly people. This reverse adaptive logic is completely absent in traditional single fixed-speed speech systems.

[0035] By collaborating across multiple modules—collection, recognition, weighted association, and speech synthesis—a cross-module adaptive compensation mechanism is formed. The more unclear the elderly person's pronunciation and the more obvious their expression difficulties, the higher the auditory compensation level of the synthesized speech. This improves the accuracy of fuzzy speech recognition while simultaneously enhancing the elderly person's comprehension of the broadcast speech. Using fuzzy recognition or elderly speech synthesis alone can only solve a single problem; however, this invention, through module linkage and dual-formula association, achieves dynamic matching between the degree of recognition degradation and the intensity of synthesis compensation. The worse the elderly person's communication conditions, the stronger the system's assistance capability, resulting in improved overall communication efficiency. This solves a synergistic effect that cannot be achieved even with the combined use of existing technologies, demonstrating significant innovation.

[0036] The terminal interaction module is divided into a local interaction terminal for the elderly and a remote interaction terminal for caregivers. The local interaction terminal for the elderly features large buttons, a high-contrast display interface, and extra-large fonts, supporting one-click voice input and minimalist operation. The remote interaction terminal for caregivers can be any of a mobile phone, tablet, or computer, receiving standard semantic text converted from the elderly's ambiguous speech in real time, supporting remote text / voice replies from family members, caregivers, and medical staff, achieving two-way barrier-free interaction in elderly care scenarios. The above defines a layered terminal architecture. The minimalist design of the local terminal for the elderly is suitable for elderly people with declining vision, slow operation, and weakened learning ability; the remote terminal for caregivers is mainly for text viewing, quickly understanding the elderly's ambiguous requests, providing remote replies and real-time care, suitable for scenarios such as remote elderly care by children, shift care by caregivers, and collaborative care between medical staff in wards.

[0037] Example as follows: Unified fixed parameters: Example 1: Severe speech impairment scenario (elderly person with cerebral infarction, whose speech is intermittent and unclear) The elderly person's account: Head...dizziness...feeling unwell...want to lie down Split into segments: 1. Parameters for each segment 2. Formula 1: Piecewise Calculation Semantic score: 3. Calculate the average weight. 4. Formula 2 Prosody Calculation and Simplification Standard text: "I feel dizzy and unwell, and I want to lie down and rest."

[0038] Standard text after editing: "I feel dizzy and unwell, and I want to lie down and rest," character length 5. Actual Results High compensation value: speech rate reduced to 0.6 times, sentence pauses to 0.6 seconds, low frequency enhancement, perfectly suited for elderly people with severe speech impairment.

[0039] Example 2: Mild aging speech scenario (young, healthy elderly person, speaking clearly) What I said: I feel fine today and don't need to take any medicine. The calculation yields: Text length Substitute into the formula The numerical value is significantly reduced, the synthesized speech rhythm is slow but not sluggish, close to the normal elderly-friendly speech speed, and there is no overcompensation.

[0040] Two sets of examples demonstrate that the two sets of formulas are closely related, have rigorous calculation logic, and have obvious positive and negative contrasts. The technical effect of stronger compensation with higher ambiguity can be quantified and reproduced.

[0041] A method for implementing synthesized speech for the elderly based on elderly care, applied to the aforementioned system, includes the following steps: Step S1: An elderly fuzzy speech acquisition and preprocessing module acquires the original speech signal of the elderly person with unclear pronunciation and fragmented speech in real time, performs noise reduction, volume gain, and sentence segment splicing preprocessing, and outputs an optimized and clean speech signal; Step S2: An elderly fuzzy speech recognition and semantic sorting module extracts acoustic features from the optimized speech and matches it with an elderly care semantic database, substitutes it into a semantic correction formula to complete quantitative correction, outputs standard and fluent semantic text, and simultaneously generates semantic weight coefficients corresponding to each speech segment; Step S3: A semantic weight dynamic association module... The system collects all semantic weight coefficients, calculates the average semantic weight coefficient, and encrypts and pushes this associated parameter to the elderly-adaptive speech synthesis module in real time. Step S4: The elderly-adaptive speech synthesis module retrieves the average semantic weight coefficient, substitutes it into the prosodic control formula to calculate the prosodic comprehensive control compensation value, and adaptively adjusts the speech rate, pauses, and low-frequency gain parameters based on the compensation value to generate customized synthesized speech adapted to the auditory characteristics of the elderly. Step S5: The terminal interaction module pushes the standard semantic text to the caregiver's remote interaction terminal, while simultaneously completing gentle, adapted voice playback on the elderly person's local interaction terminal. This process repeats continuously, completing a two-way closed-loop voice interaction between the elderly person and their caregiver. The above is the complete method flow corresponding to the system, with closed-loop steps and coherent logic. From voice acquisition, preprocessing, semantic formula calculation, weight averaging linkage, prosodic formula control, voice playback, and text push, the entire process is automated, requiring no complex operations from the elderly person. Two-way interaction can be triggered with a single click. The method can be implemented through software programming and is compatible with various embedded elderly care hardware.

[0042] Working principle of the invention: When an elderly person initiates a voice interaction at close range, the elderly fuzzy speech acquisition and preprocessing module collects the original weak speech. After noise reduction, gain, and segment splicing optimization, it is sent to the elderly fuzzy speech recognition and semantic sorting module. The recognition module, relying on an elderly-specific dataset, uses a semantic correction formula to complete the semantic repair of incomplete speech and fuzzy pronunciation, outputting fluent standard text and the semantic weight coefficient of each speech segment. The semantic weight dynamic association module integrates all weights, calculates the average semantic weight coefficient, and sends it to the elderly-adaptive speech synthesis module in real time. The synthesis module receives the linkage parameters, substitutes them into the reverse prosodic control formula, calculates the prosodic compensation value, and adaptively adjusts the speech rate, pauses, and timbre frequency bands to generate customized elderly-adaptive speech. Finally, the terminal interaction module sends the repaired standard text to family members, caregivers, and medical staff terminals, and plays the gentle and slow synthesized speech on the elderly's terminal, forming a two-way closed-loop interaction of "the elderly's fuzzy expression, clear translation, and customized voice response." Relying on module linkage and dual-formula quantification constraints, it achieves unexpected technical effects of adaptive auditory compensation.

[0043] All the logic of this invention can be realized through the combination of software algorithms, embedded chips, and smart terminal hardware for the elderly. It can be mass-produced and installed in elderly wristbands, bedside elderly care terminals, elderly care speakers, and ward interactive devices. It is applicable to all types of elderly care scenarios and has strong practicality and industrial promotion value.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for synthesizing speech for the elderly based on elderly care, characterized in that, include: The system comprises a fuzzy speech acquisition and preprocessing module for the elderly, a fuzzy speech recognition and semantic analysis module for the elderly, a dynamic semantic weight association module, an elderly-adapted speech synthesis module, and a terminal interaction module. The fuzzy speech acquisition and preprocessing module acquires raw speech signals from elderly individuals with unclear pronunciation, ambiguous articulation, and discontinuous sentences, and performs noise reduction, volume gain, and fragmented segment splicing preprocessing on the raw speech signals. The fuzzy speech recognition and semantic analysis module, communicatively connected to the fuzzy speech acquisition and preprocessing module, extracts acoustic features from the preprocessed speech signals, performs elderly-specific semantic matching, corrects fragmented and pronunciation-deviation-prone speech expressions, and outputs standardized and accurate semantic text along with semantic weight coefficients for each speech segment. The dynamic semantic weight association module is electrically connected to the fuzzy speech recognition module for the elderly. Together with the semantic analysis module and the age-adaptive speech synthesis module, it receives the semantic weight coefficients of all speech segments, calculates and generates the average semantic weight coefficient, and transmits the average semantic weight coefficient to the age-adaptive speech synthesis module in real time, realizing the quantitative linkage between the two major functional modules of recognition and synthesis. The age-adaptive speech synthesis module receives the average semantic weight coefficient issued by the semantic weight dynamic association module, and uses it to adaptively adjust the speech prosody parameters in combination with the average semantic weight coefficient, converting the standard text or standard speech input from the outside into customized synthesized speech adapted to the hearing degeneration characteristics of the elderly. The terminal interaction module establishes data interaction with each of the aforementioned modules, and is used to realize the elderly's voice input, local broadcast of age-adaptive synthesized speech, standardized semantic text visualization display, and two-way information transmission and interaction between multiple terminals.

2. The system for synthesizing speech for the elderly based on elderly care, as described in claim 1, is characterized in that: The elderly fuzzy speech acquisition and preprocessing module has a built-in adaptive noise reduction unit specifically designed for elderly care scenarios. It targets the characteristics of elderly people's low speaking volume, a lot of oral airflow noise, and background interference in home / elderly care institutions. It filters out invalid environmental noise and human voice noise in a targeted manner, improving the signal-to-noise ratio of the acquired speech signal to more than 25dB. At the same time, it fully preserves the core low-frequency acoustic features of the elderly's voice, avoiding the loss of effective fuzzy speech features.

3. The system for synthesizing speech for the elderly based on elderly care, as described in claim 1, is characterized in that: The elderly fuzzy speech recognition and semantic sorting module has a built-in fuzzy speech dataset and semantic correction model specifically for the elderly. The semantic correction model is configured with a semantic correction formula for elderly fuzzy speech, and completes the semantic correction and integration of fragmented fuzzy speech through quantitative calculation. The semantic correction formula for ambiguous speech in the elderly is as follows: ;in, The overall score is based on semantic correction. The total number of independent speech segments for decomposing ambiguous speech in the elderly. For the first The semantic weight coefficients of each speech segment. For the original acoustic feature values ​​of fuzzy speech, Matching score for a semantic database specifically designed for elderly care; These are the speech feature weighting coefficients, with a value range of... This refers to the semantic matching weight coefficient, with a value range of... This is the noise interference penalty coefficient, with a value range of... This is the quantized value of the noise features corresponding to a single speech segment.

4. A system for synthesizing speech for the elderly based on elderly care, as described in claim 3, characterized in that: The semantic weight coefficient The system comprehensively quantifies and judges based on the pronunciation completeness, articulation clarity, and sentence coherence of a single speech segment. The range of values ​​is ; The more unclear the pronunciation and the more severe the speech breaks of an elderly person, the higher the semantic weight coefficient of the corresponding segment. The smaller the value, the clearer and more complete the elderly person's pronunciation and expression, and the higher the semantic weight coefficient of the corresponding segment. The larger the value.

5. A system for synthesizing speech for the elderly based on elderly care, as described in claim 3, characterized in that: The semantic weight dynamic association module retrieves the semantic weight coefficients of all speech segments. The average semantic weight coefficient is calculated using the arithmetic mean algorithm. ; ； The average semantic weight coefficient As a core quantization parameter for cross-module association, it is sent unidirectionally and in real time to the elderly-adaptive speech synthesis module, realizing closed-loop linkage from semantic recognition results to the speech synthesis control end.

6. A system for synthesizing speech for the elderly based on elderly care, as described in claim 5, characterized in that: The elderly-adaptive speech synthesis module incorporates an elderly-adaptive speech prosody control model. This model is configured with an elderly-adaptive speech prosody control formula, which, combined with the received average semantic weight coefficients, performs adaptive quantization adjustment of speech rate, pauses, and pitch. The elderly-adaptive speech prosody control formula is as follows: in, This is the comprehensive prosody regulation compensation value. The average semantic weight coefficient output by the semantic weight dynamic association module; Fixed coefficients are used to fit hearing to older adults, with a range of values. This is a benchmark value for standard speaking speed; This is a statement length control coefficient, with a value range of... The length of the statement in the text to be synthesized.

7. A system for synthesizing speech for the elderly based on elderly care, as described in claim 6, characterized in that: The rhythmic comprehensive regulation compensation value With average semantic weight coefficient A negative correlation mapping relationship exists; average semantic weight coefficient The smaller the value, the better the overall rhythm regulation compensation value. The larger the average semantic weight coefficient, the faster the synthesized speech rate, the longer the pause intervals, the stronger the low-frequency human voice gain, and the weaker the high-frequency sharp timbre; The larger the value, the greater the overall rhythm regulation compensation value. The smaller the volume, the more the synthesized speech will restore the normal speaking speed and natural rhythm, avoiding sluggish broadcasting.

8. A system for synthesizing speech for the elderly based on elderly care, as described in claim 1, characterized in that: By coordinating multiple modules such as data collection, recognition, weighted association, and speech synthesis, a cross-module adaptive compensation mechanism is formed. The more unclear the elderly person's pronunciation and the more obvious the expression impairment, the higher the auditory compensation level of the system for the feedback synthesized speech. While improving the accuracy of fuzzy speech recognition, it also improves the elderly person's ability to understand the broadcast speech.

9. A system for synthesizing speech for the elderly based on elderly care, as described in claim 1, characterized in that: The terminal interaction module is divided into a local interaction terminal for the elderly and a remote interaction terminal for caregivers. The local interaction terminal for the elderly is equipped with large buttons, a high-contrast display interface, and extra-large fonts, and supports one-click voice input and extremely simple operation. The remote interaction terminal for caregivers can be any of a mobile phone, tablet, or computer, and can receive standard semantic text converted from the elderly's fuzzy speech in real time. It supports remote text / voice replies from family members, caregivers, and medical staff, realizing two-way barrier-free interaction in elderly care scenarios.

10. A method for implementing synthesized speech for the elderly based on elderly care, characterized in that, The system for synthesized speech for the elderly based on any one of claims 1 to 9 includes the following steps: Step S1: The elderly fuzzy speech acquisition and preprocessing module acquires the original speech signal of the elderly with unclear pronunciation and fragmented speech in real time, performs noise reduction, volume gain, and sentence segment splicing preprocessing, and outputs the optimized pure speech signal; Step S2: The elderly fuzzy speech recognition and semantic sorting module extracts acoustic features and matches them with the elderly semantic database, substitutes them into the semantic correction formula to complete the quantitative correction, outputs standard and fluent semantic text, and simultaneously generates the semantic weight coefficients corresponding to each speech segment; Step S3: Semantic weights are dynamically associated. The module collects all semantic weight coefficients, calculates the average semantic weight coefficient, and encrypts and pushes this associated parameter to the elderly-adaptive speech synthesis module in real time; Step S4: The elderly-adaptive speech synthesis module retrieves the average semantic weight coefficient, substitutes it into the prosody control formula to calculate the prosody comprehensive control compensation value, and adaptively adjusts the speech rate, pauses, and low-frequency gain parameters according to the compensation value to generate customized synthesized speech adapted to the auditory characteristics of the elderly; Step S5: The terminal interaction module pushes the standard semantic text to the caregiver's remote interaction terminal, and at the same time completes the gentle and adapted speech broadcast on the elderly's local interaction terminal, repeating the process to continuously complete the two-way closed-loop voice interaction between the elderly and the caregiver.

Images