A robot human-simulated voice generation method, device, equipment and storage medium

Abstract

The application discloses a robot human-simulated voice generation method, device and equipment and a storage medium. The method is applied to a robot with a tongue speaker and comprises the following steps: obtaining input data, generating an acoustic feature sequence according to the input data; synthesizing an initial human-simulated voice signal based on the acoustic feature sequence; compensating the initial human-simulated voice signal to generate a compensated voice signal; and outputting the compensated voice signal to the tongue speaker for playing. The generation of the acoustic feature sequence can realize multi-dimensional parameterization control, completely retain pronunciation content, rhythm and intonation style, and make the voice closer to the human voice law. The synthesis of the initial human-simulated voice can simulate the human voice sound mechanism, fuse rhythm and breath features, and reduce the mechanical feeling of the voice. The compensation can reduce the sound distortion caused by the tongue material and the oral sound path, and guarantee the sound quality restoration degree. Finally, the tongue speaker is played, the mouth sound source is sounded, and the humanization degree and the interactive naturalness of the robot voice are improved.

Description

Technical Field

[0001] This invention relates to the field of speech signal processing, and in particular to a method, apparatus, device, and storage medium for generating humanoid speech for robots. Background Technology

[0002] With the continuous development of humanoid robot technology, the requirements for the degree of anthropomorphism in human-computer interaction are constantly increasing. As the core interaction method of humanoid robots, the naturalness of voice and the realism of the sound source directly determine the user's interaction experience.

[0003] In current service robot voice generation solutions, speakers are typically placed in the chest cavity, head, or neck, and synthesized speech is directly driven by the speakers. At the speech generation algorithm level, ordinary text-to-speech or voice-changing algorithms mainly adjust pitch, timbre, and speech rate, rarely simultaneously simulating stress, unstressed sounds, pauses, intonation, periodic vocal cord vibration characteristics, and breathing characteristics. For scenarios with built-in tongue speakers, current practices often involve directly inserting ordinary miniature speakers into the tongue and using general voice-driven solutions, without designing a corresponding speech generation and correction process specifically for the unique acoustic path formed by the tongue and oral cavity.

[0004] Existing technologies have speaker positions that deviate from the human mouth, easily leading to an unnatural sound from the robot and failing to achieve a highly human-like sound source location. General-purpose speech algorithms lack complete simulation of human voice characteristics, resulting in output speech that sounds mechanical and lacks naturalness. Furthermore, the narrow space of the tongue, the absorption of high frequencies by flexible materials, and the altered frequency response of the robot's oral cavity can cause the directly output sound to sound muffled, with frequency distortion and discrepancies. Simply adjusting the speaker position or performing only general speech processing cannot fully meet the high-fidelity and high-naturalness human-like vocalization requirements of humanoid robots. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for generating humanoid speech in robots, which solves the technical problems of low high-frequency attenuation, auditory distortion, and low naturalness and fidelity of speech caused by tongue absorption and oral cavity resonance in humanoid robots equipped with tongue speakers.

[0006] According to one aspect of the present invention, a method for generating humanoid speech in a robot is provided, applicable to a robot equipped with a tongue-shaped speaker, the method comprising:

[0007] Acquire input data and generate an acoustic feature sequence based on the input data;

[0008] Initial anthropomorphic speech signal was synthesized based on acoustic feature sequences;

[0009] The initial anthropomorphic speech signal is compensated to generate a compensated speech signal;

[0010] The compensated speech signal is output to the tongue speaker for playback.

[0011] Optionally, the input data includes text data. The acoustic feature sequence is generated based on the input data, including: performing text processing and semantic recognition on the text data to generate linguistic features; determining the prosodic parameters and frame-level acoustic parameters corresponding to the linguistic features through a preset prosodic template or prosodic prediction model; and temporally aligning the linguistic features, prosodic parameters, and frame-level acoustic parameters to generate the acoustic feature sequence.

[0012] Optionally, the input data includes reference speech. Based on the input data, an acoustic feature sequence is generated, including: performing frame segmentation on the reference speech to distinguish between spoken and silent frames; extracting fundamental frequency values ​​and frame-level acoustic parameters from the spoken frames; performing phoneme recognition on the reference speech to obtain linguistic features, and combining the fundamental frequency values ​​to generate prosodic parameters and linguistic features; and temporally aligning the linguistic features, prosodic parameters, and frame-level acoustic parameters to generate an acoustic feature sequence.

[0013] Optionally, the acoustic feature sequence consists of time-aligned linguistic features, prosodic parameters, and frame-level acoustic parameters. The linguistic features include phonemes and tones, the prosodic parameters include duration, stress level, pauses, fundamental frequency trajectory, and amplitude envelope, and the frame-level acoustic parameters include spectral envelope, periodic energy ratio, and voice style parameters.

[0014] Optionally, the initial anthropomorphic speech signal is synthesized based on the acoustic feature sequence, including: generating a mixed speech excitation signal according to the amplitude envelope, fundamental frequency trajectory and periodic energy ratio, wherein the mixed speech excitation includes voiced excitation and unvoiced excitation; constructing duct filtering parameters according to the spectral envelope, performing duct filtering processing on the mixed speech excitation signal through the duct filtering parameters to obtain the filtered speech signal; and using the filtered speech signal as the initial anthropomorphic speech signal.

[0015] Optionally, the initial anthropomorphic speech signal is compensated to generate a compensated speech signal, including: obtaining pre-calibrated acoustic compensation parameters, wherein the acoustic compensation parameters are constructed based on the actual acoustic path frequency response formed by the tongue speaker, tongue material, and robotic oral cavity; transforming the initial anthropomorphic speech signal to the frequency domain to obtain the initial speech spectrum; equalizing, gaining, and adjusting the dynamic range of the initial speech spectrum according to the acoustic compensation parameters to obtain the compensated speech spectrum; and inversely transforming the compensated speech spectrum into a time domain signal to generate the compensated speech signal.

[0016] Optionally, the compensated speech signal is output to the tongue speaker for playback, including: performing safe amplitude limiting processing on the compensated speech signal based on a preset amplitude range corresponding to the rated power of the tongue speaker to obtain a limited time-domain signal; performing digital-to-analog conversion on the limited time-domain signal to obtain an analog drive signal; and inputting the analog drive signal into the drive circuit of the tongue speaker to drive the tongue speaker to complete the speech playback.

[0017] According to another aspect of the present invention, a robot-inspired human voice generation device is provided, the device comprising:

[0018] The acoustic feature sequence generation module is used to acquire input data and generate acoustic feature sequences based on the input data.

[0019] The initial signal synthesis module is used to synthesize an initial anthropomorphic speech signal based on acoustic feature sequences;

[0020] The voice signal compensation module is used to compensate the initial humanoid voice signal for frequency response and distortion based on the preset acoustic compensation parameters of the robot's tongue speaker, and generate a compensated voice signal.

[0021] The voice signal playback module is used to output the compensated voice signal to the tongue speaker for playback.

[0022] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0023] At least one processor;

[0024] and a memory communicatively connected to the at least one processor;

[0025] The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform a robot-inspired human speech generation method according to any embodiment of the present invention.

[0026] According to another aspect of the present invention, a computer storage medium is provided, the computer storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement a robot-inspired human voice generation method according to any embodiment of the present invention.

[0027] The technical solution of this invention enables multi-dimensional parameterized control by generating acoustic feature sequences, fully preserving the pronunciation content, rhythm, and timbre style, making the voice closer to the natural sounds of the human voice. By synthesizing initial anthropomorphic speech, simulating the mechanism of human vocalization, and integrating rhythm and breath characteristics, the mechanical feel of the speech is effectively reduced. Compensation reduces sound distortion caused by tongue material and oral cavity acoustic pathways, ensuring high sound quality reproduction. Finally, the sound is played through a tongue speaker, realizing mouth-based sound source production, thus improving the anthropomorphism and naturalness of the robot's speech.

[0028] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of a robot humanoid speech generation method according to Embodiment 1 of the present invention;

[0031] Figure 2 This is a flowchart of another robot-inspired human voice generation method provided in Embodiment 2 of the present invention;

[0032] Figure 3 This is a schematic diagram of the structure of a robot-like human voice generation device according to Embodiment 3 of the present invention;

[0033] Figure 4 This is a schematic diagram of the structure of an electronic device that implements a robot-inspired human voice generation method according to an embodiment of the present invention. Detailed Implementation

[0034] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0035] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0036] Example 1

[0037] Figure 1 This is a flowchart illustrating a method for generating humanoid speech in a robot, as provided in Embodiment 1 of the present invention. This embodiment can be applied to robots equipped with tongue-shaped speakers. Figure 1 As shown, the method includes:

[0038] S110. Obtain input data and generate an acoustic feature sequence based on the input data.

[0039] Input data refers to the original input content used to generate human-like speech. It is the information source of the entire speech generation process and specifically includes the text data to be broadcast, upstream synthesized speech, real human reference speech, or the business interaction content that the robot needs to express in the current interaction scenario. Acoustic feature sequence refers to a multi-dimensional parameter sequence aligned with phonemes, syllables, or short-time frames of speech, including phonemes, Mandarin tones, syllable duration, stress levels, prosodic pause boundaries, fundamental frequency trajectory, amplitude envelope, periodic energy ratio, spectral envelope, and voice style parameters. This sequence can be extracted from real human reference speech or generated from text using prosodic templates or prosodic prediction models.

[0040] Optionally, the acoustic feature sequence consists of time-aligned linguistic features, prosodic parameters, and frame-level acoustic parameters. The linguistic features include phonemes and tones, the prosodic parameters include duration, stress level, pauses, fundamental frequency trajectory, and amplitude envelope, and the frame-level acoustic parameters include spectral envelope, periodic energy ratio, and voice style parameters.

[0041] Among them, linguistic features describe the basic identifier parameters of speech articulation units and pitch categories. These parameters serve as the temporal alignment anchors for all prosody and acoustic parameters, determined solely by the language content and unaffected by changes in prosody or timbre style. Phonemes are the smallest articulation units of speech, categorized into different types such as initials and finals, used to mark specific pronunciation content. In text input scenarios, they are obtained through phoneme conversion. In reference speech input scenarios, they are obtained through phoneme recognition. All parameters such as duration, stress, fundamental frequency, and spectrum are aligned temporally based on phonemes as the fundamental unit. Tones refer to the tone category identifiers of Chinese syllables, including high level tone, rising tone, falling-rising tone, and falling tone. They determine the local tone trend of the fundamental frequency trajectory and are the basis for generating tone components within the fundamental frequency trajectory. In text input scenarios, they are obtained through tone annotation; in reference speech input scenarios, they are obtained through tone recognition. These are linguistic-level pitch category attributes, not frame-level continuous pitch curves.

[0042] Among them, prosodic parameters refer to the temporal attributes that characterize the speed, pitch, stress, and volume of speech, determining the natural expression logic and auditory rhythm of speech. Duration refers to the length of time a single phoneme or syllable lasts. Stress level is used to mark the degree of emphasis on a syllable or phrase, determined by the contextual semantics and business focus, not by a fixed value assigned to a single Chinese character. A higher stress level corresponds to a higher duration and intensity of the syllable, reflecting the information focus of the sentence. Pauses, also known as prosodic boundaries, are set at punctuation marks, phrase boundaries, and sentence boundaries, and are divided into different levels such as short pauses and sentence-ending pauses, used to divide the rhythm of sentences and match the phrasing and breathing patterns of natural human speech. The fundamental frequency trajectory is a continuous curve showing the change of the fundamental frequency of the voiced segments of speech over time. The fundamental frequency trajectory determines the overall pitch trend of speech, carrying the pitch expression of word intonation, sentence intonation, and stress. Amplitude envelope is the envelope curve of the overall energy of speech over time. It controls the overall volume fluctuation of speech and works in conjunction with stress level and pauses. The amplitude of stressed areas increases and the amplitude of pause areas approaches zero, together reflecting the rhythm of the speech.

[0043] Among them, frame-level acoustic parameters refer to fine-grained acoustic parameters aligned frame-by-frame according to short-time speech, controlling the timbre and the proportion of vocal components, and determining the final auditory quality and stylistic attributes of the voice. The spectral envelope is a parameter that characterizes the overall shape of the speech spectrum, corresponding to the vocal tract filtering characteristics of human vocalization. The periodic energy ratio ranges from 0 to 1, used to control the energy ratio of periodic glottal components and non-periodic components. This value is higher for voiced segments such as vowels, lower for voiceless consonants, and appropriately reduced for breathy voice styles. Voice style parameters are used to regulate the overall style attributes of the voice, such as approachable, steady, youthful, and neutral. Voice style parameters are independent of the language content; under the same text and prosody, they can output different styles of voice auditory perception, belonging to global acoustic style control parameters.

[0044] For example, the acoustic feature sequence can be represented by the following formula (1):

[0045] (1)

[0046] in, Representing phonemes, Indicates tone, Indicates duration, Indicates the degree of accent. Indicates a pause. Indicates the fundamental frequency trajectory. Indicates the amplitude envelope. Indicates the periodic energy ratio, Represents the spectral envelope. This represents vocal style parameters.

[0047] Optionally, the input data includes text data. The acoustic feature sequence is generated based on the input data, including: performing text processing and semantic recognition on the text data to generate linguistic features; determining the prosodic parameters and frame-level acoustic parameters corresponding to the linguistic features through a preset prosodic template or prosodic prediction model; and temporally aligning the linguistic features, prosodic parameters, and frame-level acoustic parameters to generate the acoustic feature sequence.

[0048] Specifically, text processing includes operations such as text normalization, word segmentation, phoneme conversion, and Mandarin tone annotation. Semantic recognition is responsible for identifying punctuation, sentence structure, business interaction intent, and semantic focus in the text. The resulting linguistic features are the basic identifier parameters describing speech articulation units and pitch categories, including phonemes and tones. Pre-set prosodic templates are suitable for standardized dialogue in fixed scenarios, and pre-made prosodic rules and parameter values ​​can be directly reused. The prosodic prediction model is used to process dynamic text and can automatically output corresponding parameters based on linguistic features. For fixed dialogue, all the above parameters can be directly extracted from real-person reference recordings. For dynamic text, the model performs prediction, and parameter values ​​can be adjusted based on business keywords or dialogue status. Temporal alignment refers to matching the three types of parameters according to the actual pronunciation time sequence, in units of phonemes, syllables, or short frames, to form a complete acoustic feature sequence.

[0049] Optionally, the input data includes reference speech. Based on the input data, an acoustic feature sequence is generated, including: performing frame segmentation on the reference speech to distinguish between spoken and silent frames; extracting fundamental frequency values ​​and frame-level acoustic parameters from the spoken frames; performing phoneme recognition on the reference speech to obtain linguistic features, and combining the fundamental frequency values ​​to generate prosodic parameters and linguistic features; and temporally aligning the linguistic features, prosodic parameters, and frame-level acoustic parameters to generate an acoustic feature sequence.

[0050] Specifically, frame segmentation refers to dividing a continuous speech signal into multiple short-time frames according to a fixed-duration short-term window, adapting to the short-term stationary characteristics of speech signals to facilitate frame-by-frame analysis. Spoken frames correspond to periods with effective vocalizations such as voiced and voiceless consonants. Silent frames correspond to periods of pauses and silence. After differentiation, parameter extraction can be performed on effective vocal segments, avoiding the introduction of invalid data from silent segments. The fundamental frequency value refers to the fundamental frequency value corresponding to the voiced sounds in each frame, which is the basic data for constructing the fundamental frequency trajectory. Through phoneme recognition and tone recognition, the pronunciation content and tone category corresponding to each time period in the speech are clearly identified, and complete prosodic parameters are generated by combining prosodic annotation and frame-by-frame fundamental frequency values. Temporal alignment involves matching the three types of parameters according to the actual timeline of the reference speech, using phonemes, syllables, or short time frames as units. This ensures that the linguistic attributes, prosodic expression, and acoustic details of each pronunciation unit correspond completely in the time dimension. The resulting acoustic feature sequence retains the natural prosody and timbre of the reference speech while also possessing adjustable parameterization characteristics, which can be directly used for subsequent synthesis of human-like voice signals.

[0051] In one specific implementation, the fundamental frequency trajectory of the acoustic frame can be represented by the following formula (2):

[0052] (2)

[0053] in, Indicates the fundamental frequency trajectory. The reference pitch representing the target voice. This represents the tone curve of Mandarin Chinese. Indicates the intonation of phrases and sentences. Indicates local changes caused by stress or semantic focus. It indicates the effects of tone sandhi, linking, and adjacent syllables.

[0054] For example, consider the application scenario of a hotel service robot equipped with a tongue speaker broadcasting "Welcome, do you need to check in?". In the acoustic feature sequence generation stage, the text to be broadcast is first divided into multiple speech segments. Prosody and stress rules are set based on the hotel service inquiry context: "Welcome" is set to normal or secondary stress, a short pause is set at the comma position, "Do you need to check in?" is set as a polite introductory word and is read lightly, "Check in?" is the core information area, with "Check in?" set as the semantic focus and its duration appropriately extended and its intensity enhanced, and "Do you need to check in?" set as a light reading with an added interrogative intonation boundary. The stress setting is dynamically determined by the contextual semantics, not based on a fixed lookup table for a single Chinese character. If there is a semantic contrast between "need" and "don't need" in the preceding text, the main stress can be adjusted to "need". If there is a semantic contrast between "Check in?" and "Extend your stay", the main stress falls on "Check in?".

[0055] S120. Synthesize the initial anthropomorphic speech signal based on acoustic feature sequences.

[0056] The initial anthropomorphic speech signal refers to the initial time-domain speech waveform synthesized based on acoustic feature sequences, i.e., the human-like voice time-domain signal. During synthesis, the fundamental frequency trajectory controls the vibration period of the periodic glottal component, the amplitude envelope and periodic energy ratio control the amplitude proportions of voiced and breath / voiceless consonants, and the spectral envelope parameter controls the final timbre and vocal tract characteristics. The output signal already conforms to the vocal production rules of the human voice, possessing natural stress, intonation, and breathiness, but it has not yet adapted to the acoustic path characteristics of the tongue speaker and the robotic mouth. Specific waveforms can be generated by a parametric synthesizer, a neural vocoder, a text-to-speech model, or a controlled voice-changing model.

[0057] Optionally, the initial anthropomorphic speech signal is synthesized based on the acoustic feature sequence, including: generating a mixed speech excitation signal according to the amplitude envelope, fundamental frequency trajectory and periodic energy ratio, wherein the mixed speech excitation includes voiced excitation and unvoiced excitation; constructing duct filtering parameters according to the spectral envelope, performing duct filtering processing on the mixed speech excitation signal through the duct filtering parameters to obtain the filtered speech signal; and using the filtered speech signal as the initial anthropomorphic speech signal.

[0058] Specifically, the amplitude coefficients of the periodic and non-periodic components are first determined based on the amplitude envelope and the ratio of periodic energy, respectively. The vibration period of the periodic component is then controlled by the fundamental frequency trajectory to generate a mixed speech excitation signal containing voiced and unvoiced excitations. The voiced excitation corresponds to the periodic glottal component, primarily carrying the vocal energy of voiced segments such as vowels; the unvoiced excitation corresponds to the non-periodic random excitation component, used to simulate the friction, plosive effects, and natural breathiness of voiceless consonants.

[0059] For example, the voiced excitation amplitude coefficient is shown in the following formula (3):

[0060] (3)

[0061] in, This represents the amplitude coefficient of the voiced sound excitation. Indicates the amplitude envelope. This represents the periodic energy ratio. The amplitude coefficient of the clear tone excitation is given by the following formula (4):

[0062] (4)

[0063] in, This represents the amplitude coefficient of the unvoiced tone excitation. Indicates the amplitude envelope. Indicates the periodic energy ratio, The values ​​range from 0 to 1. The periodic energy ratio of voiced segments such as vowels is relatively high, corresponding to voiced excitation being dominant. The periodic energy ratio of voiceless consonant segments is relatively low, corresponding to voiceless excitation being dominant. The amplitudes of both types of excitation in pause segments are close to zero. After weighting and superimposing the two types of excitation according to their corresponding amplitudes, the complete mixed speech excitation signal can be obtained, as shown in the following formula (5):

[0064] (5)

[0065] in, This represents a mixed speech excitation signal. This represents the amplitude coefficient of the voiced sound excitation. This represents the amplitude coefficient of the unvoiced tone excitation. Indicates the fundamental frequency trajectory. This represents the periodic glottal component that varies with the fundamental frequency trajectory. This represents the non-periodic excitation component. Then, based on the spectral envelope, the vocal tract filtering parameters are determined, and the mixed speech excitation signal is subjected to vocal tract filtering to obtain the filtered speech signal. The spectral envelope is a frame-level parameter that characterizes the overall shape of the speech spectrum, corresponding to the vocal tract filtering characteristics in the human vocalization process. The vocal tract filtering parameters are digital filtering models constructed based on the spectral envelope, used to simulate the resonant shaping effect of the human oral cavity, pharynx, and other vocal tract structures on the sound source. After vocal tract filtering, the excitation signal will be endowed with the formant features of the corresponding timbre, and at the same time, the voice style parameters can be integrated into the filtering process by adjusting the shape of the spectral envelope to achieve the control of different voice styles. The filtered speech signal is represented by the following formula (6):

[0066] (6)

[0067] in, This represents the filtered speech signal. This refers to duct filtering or vocoder processing based on the spectral envelope, used to simulate the shaping effect of the vocal tract on the sound source. The filtering process endows the excitation signal with the formant and timbre attributes of the target voice. At the same time, voice style parameters can be integrated into the filtering process by adjusting the shape of the spectral envelope, thereby achieving the control of different voice styles such as friendly, steady, and youthful. This represents the mixed speech excitation signal. The filtered speech has fully integrated the pitch pattern corresponding to the fundamental frequency, the vocal components corresponding to the periodic energy ratio, the timbre characteristics corresponding to the spectral envelope, and the rhythmic variations corresponding to the amplitude envelope. It completely conforms to the vocal mechanism of the natural human voice and can be directly fed into the subsequent acoustic path compensation stage.

[0068] S130. Compensate the initial anthropomorphic speech signal to generate a compensated speech signal.

[0069] The acoustic path compensation refers to the correction process for the sound production path comprised of the tongue speaker, the flexible tongue material, and the robotic oral cavity. First, the actual frequency response of the entire acoustic path ("speaker-tongue-oral cavity-mouth") is measured using factory calibration. Then, a compensation filter with stable constraints is constructed to equalize, gain, and adjust the dynamic range of the initial anthropomorphic speech signal's spectrum. This compensates for issues such as muffled sound and frequency distortion caused by the tongue material absorbing high frequencies and the oral cavity altering its frequency response, ensuring the fidelity of the sound output from the mouth. This step only corrects acoustic path deviations and does not alter the pitch, stress, intonation, or vocal style of the speech.

[0070] Optionally, the initial anthropomorphic speech signal is compensated to generate a compensated speech signal, including: obtaining pre-calibrated acoustic compensation parameters, wherein the acoustic compensation parameters are constructed based on the actual acoustic path frequency response formed by the tongue speaker, tongue material, and robotic oral cavity; transforming the initial anthropomorphic speech signal to the frequency domain to obtain the initial speech spectrum; equalizing, gaining, and adjusting the dynamic range of the initial speech spectrum according to the acoustic compensation parameters to obtain the compensated speech spectrum; and inversely transforming the compensated speech spectrum into a time domain signal to generate the compensated speech signal.

[0071] Specifically, firstly, pre-calibrated acoustic compensation parameters are obtained. These parameters are constructed based on the actual acoustic path frequency response formed by the tongue speaker, tongue material, and robot mouth. The actual acoustic path frequency response is the overall acoustic path transmission characteristic measured during the factory calibration phase. The calibration direction is used to input the standard calibration signal to the tongue speaker after installation, and at the same time, the actual output sound spectrum is collected through a reference microphone at the reference position of the robot mouth. The ratio of the two is the actual acoustic path frequency response from the speaker to the mouth. It is expressed by the following formula (7):

[0072] (7)

[0073] in, This represents the actual sound path frequency response from the tongue speaker through the tongue and oral cavity to the reference position at the mouth. This is represented as the calibration output spectrum acquired at a reference position in the robot's mouth. This represents the standard calibration signal input to the tongue speaker. The actual acoustic path frequency response directly reflects the gain deviation in each frequency band caused by factors such as the tongue material's absorption of high frequencies and oral cavity resonance. The acoustic compensation parameters are essentially compensation filter coefficients with stability constraints constructed based on this measured frequency response. A stability coefficient greater than zero is introduced to prevent the deeply attenuated frequency bands from being over-amplified when directly inverting the acoustic path response, thus introducing noise or causing hardware overload and ensuring the stability of the compensation process.

[0074] Then, the initial anthropomorphic speech signal is transformed to the frequency domain to obtain the initial speech spectrum. Since the speech signal is a time-varying signal, its spectral characteristics change dynamically with time. Therefore, during processing, the initial anthropomorphic speech signal is divided into short frames of fixed duration, and then a Fourier transform is performed on each frame signal to obtain the initial speech spectrum of the corresponding frame, as shown in the following formula (8):

[0075] (8)

[0076] in, This represents the initial speech spectrum. This represents the filtered speech signal, which is also the initial anthropomorphic speech signal.

[0077] Furthermore, the initial speech spectrum can be equalized, and the gain and dynamic range can be adjusted according to the acoustic compensation parameters to obtain the compensated speech spectrum. The equalization process is used to correct the gain deviation caused by the acoustic path on a frequency band basis, and to offset problems such as high-frequency absorption by the tongue and the protrusion of the oral cavity frequency band, so that the sound returns to the target response with a flatness. The gain adjustment is used to match the overall output level to adapt to the rated operating range of the loudspeaker. The dynamic range adjustment is used to constrain the signal peak to avoid hardware distortion. In the specific calculation, the initial speech spectrum can be multiplied by the compensation filter on a frequency point basis, as shown in the following formula (9):

[0078] (9)

[0079] in, This represents the compensated speech spectrum. This represents the initial speech spectrum. Let represent the compensation filter. The calculation of the compensation filter follows a relationship with stability constraints, as shown in the following formula (10):

[0080] (10)

[0081] in, Indicates a compensation filter. This represents the actual sound path frequency response from the tongue speaker through the tongue and oral cavity to the reference position at the mouth. Represents the complex conjugate of the actual acoustic path frequency response. This refers to the preset target transmission response, that is, the target transmission response expected to be formed from the tongue speaker to the robot's mouth, rather than the spectrum of the target speech itself. This represents a stability coefficient greater than zero, used to limit excessive gain. The desired transmission characteristics of the sound path are expressed in the following formula (11): After the compensated speech spectrum is transmitted through the tongue speaker and the oral cavity sound path, the final output spectrum at the mouth reference position is as follows:

[0082] (11)

[0083] in, The final output spectrum represents the mouth reference position. Indicates a compensation filter. This represents the actual sound path frequency response from the tongue speaker through the tongue and oral cavity to the reference position at the mouth. This represents the initial speech spectrum. Under the premise that the stability coefficient is extremely small... The product of the target response, which is very close to the preset target, and the initial speech spectrum, is... When the target transmission response value is 1, the final output sound can restore the original hearing of the initial anthropomorphic speech to the greatest extent. After correction, the compensated speech spectrum can be obtained. Finally, the compensated speech spectrum is inversely transformed into a time-domain signal to generate a compensated speech signal. That is, an inverse Fourier transform is performed on the frequency domain signal after frame-by-frame correction to convert the frequency domain spectrum data back into a discrete waveform in the time domain. Then, the time-domain waveforms of each frame are spliced ​​together to obtain a complete compensated time-domain speech signal. This signal has been fully adapted to the acoustic path characteristics of the tongue speaker and the robot's oral cavity. After driving the speaker to produce sound, it can output an anthropomorphic voice with the expected fidelity at the mouth.

[0084] S140. Output the compensated speech signal to the tongue speaker for playback.

[0085] The tongue speaker is a custom-designed miniature electroacoustic transducer installed inside the robot's tongue. It converts electrical drive signals into sound waves and is preferably installed at the base of the tongue or in the middle to rear of the tongue to preserve space for the tongue tip's movement. The sound is directed towards the robot's oral cavity, radiating outwards from the mouth opening after being emitted, thus locating the sound source close to the mouth position of a human speaker. The speaker is isolated from the tongue drive structure by elastic supports, and the connecting wire extends from the base of the tongue with allowance for bending. It should be noted that if the temperature, current, or distortion exceeds safe limits, the volume should be reduced or the sound output should be stopped.

[0086] Optionally, this application also includes a lip-sync step. The controller can send the corresponding syllable or phoneme time nodes in the acoustic feature sequence to the robot's lip, jaw, or tongue motion controller, and control the corresponding motion mechanism to perform mouth movements that match the speech content, so that the robot's lip changes are synchronized with the speech playback process, further enhancing the anthropomorphic performance of human-computer interaction. This lip-sync is an optional interaction enhancement function and is not a necessary condition for forming a human-like voice. Not performing this step will not affect the normal generation and sound production effect of human-like speech.

[0087] The technical solution of this invention enables multi-dimensional parameterized control by generating acoustic feature sequences, fully preserving the pronunciation content, rhythm, and timbre style, making the voice closer to the natural sounds of the human voice. By synthesizing initial anthropomorphic speech, simulating the mechanism of human vocalization, and integrating rhythm and breath characteristics, the mechanical feel of the speech is effectively reduced. Compensation reduces sound distortion caused by tongue material and oral cavity acoustic pathways, ensuring high sound quality reproduction. Finally, the sound is played through a tongue speaker, realizing mouth-based sound source production, thus improving the anthropomorphism and naturalness of the robot's speech.

[0088] Example 2

[0089] Figure 2 This is a flowchart of a robot-inspired humanoid speech generation method provided in Embodiment 2 of the present invention. This embodiment adds a specific process of outputting the compensated speech signal to a tongue-shaped speaker for playback, based on Embodiment 1. The specific content of steps S210-S230 is largely the same as steps S110-S130 in Embodiment 1, and therefore will not be repeated in this embodiment. Figure 2 As shown, the method includes:

[0090] S210. Obtain input data and generate an acoustic feature sequence based on the input data.

[0091] Optionally, the input data includes text data. The acoustic feature sequence is generated based on the input data, including: performing text processing and semantic recognition on the text data to generate linguistic features; determining the prosodic parameters and frame-level acoustic parameters corresponding to the linguistic features through a preset prosodic template or prosodic prediction model; and temporally aligning the linguistic features, prosodic parameters, and frame-level acoustic parameters to generate the acoustic feature sequence.

[0092] Optionally, the input data includes reference speech. Based on the input data, an acoustic feature sequence is generated, including: performing frame segmentation on the reference speech to distinguish between spoken and silent frames; extracting fundamental frequency values ​​and frame-level acoustic parameters from the spoken frames; performing phoneme recognition on the reference speech to obtain linguistic features, and combining the fundamental frequency values ​​to generate prosodic parameters and linguistic features; and temporally aligning the linguistic features, prosodic parameters, and frame-level acoustic parameters to generate an acoustic feature sequence.

[0093] Optionally, the acoustic feature sequence consists of time-aligned linguistic features, prosodic parameters, and frame-level acoustic parameters. The linguistic features include phonemes and tones, the prosodic parameters include duration, stress level, pauses, fundamental frequency trajectory, and amplitude envelope, and the frame-level acoustic parameters include spectral envelope, periodic energy ratio, and voice style parameters.

[0094] S220. Synthesize the initial anthropomorphic speech signal based on acoustic feature sequences.

[0095] Optionally, the initial anthropomorphic speech signal is synthesized based on the acoustic feature sequence, including: generating a mixed speech excitation signal according to the amplitude envelope, fundamental frequency trajectory and periodic energy ratio, wherein the mixed speech excitation includes voiced excitation and unvoiced excitation; constructing duct filtering parameters according to the spectral envelope, performing duct filtering processing on the mixed speech excitation signal through the duct filtering parameters to obtain the filtered speech signal; and using the filtered speech signal as the initial anthropomorphic speech signal.

[0096] S230. Compensate the initial anthropomorphic speech signal to generate a compensated speech signal.

[0097] Optionally, the initial anthropomorphic speech signal is compensated to generate a compensated speech signal, including: obtaining pre-calibrated acoustic compensation parameters, wherein the acoustic compensation parameters are constructed based on the actual acoustic path frequency response formed by the tongue speaker, tongue material, and robotic oral cavity; transforming the initial anthropomorphic speech signal to the frequency domain to obtain the initial speech spectrum; equalizing, gaining, and adjusting the dynamic range of the initial speech spectrum according to the acoustic compensation parameters to obtain the compensated speech spectrum; and inversely transforming the compensated speech spectrum into a time domain signal to generate the compensated speech signal.

[0098] S240: Based on the preset amplitude range corresponding to the rated power of the tongue speaker, perform safe amplitude limiting processing on the compensated speech signal to obtain the amplitude-limited time domain signal.

[0099] The compensated speech signal is a time-domain discrete digital signal after acoustic path correction. The preset amplitude range is the upper and lower limits of the signal amplitude pre-set based on the rated power, maximum withstand voltage, and linear operating range of the tongue-shaped loudspeaker. The purpose of safety limiting processing is to clamp peaks in the signal that exceed the preset amplitude range to within the range boundaries, preventing nonlinear distortion or even hardware damage to the loudspeaker due to signal power overload. Safety limiting processing ensures that the signal always operates within the linear safety range of the loudspeaker, resulting in a time-domain digital signal with compliant amplitude.

[0100] S250 converts the time-domain signal after clipping to analog to obtain an analog drive signal.

[0101] Specifically, digital-to-analog conversion is the process of converting discrete digital time-domain signals into continuously changing analog voltage signals using a digital-to-analog converter. Since speaker driver circuits cannot directly recognize and process digital signals, this step is necessary to convert digital waveforms into continuous electrical signals that can be processed by the circuit, which are analog drive signals.

[0102] S260: Input the analog drive signal into the drive circuit of the tongue speaker to drive the tongue speaker to complete the voice playback.

[0103] Specifically, the drive circuit is responsible for amplifying the input analog drive signal and outputting drive power that matches the rated parameters of the speaker. This drives the diaphragm of the tongue speaker to vibrate mechanically, converting the electrical signal into a sound wave signal. The sound wave is then transmitted through the tongue material and the cavity of the robot's mouth and radiates outward, ultimately achieving speech playback that mimics a human voice.

[0104] The technical solution of this invention uses a preset amplitude range corresponding to the rated power of the tongue speaker for safety limiting, which can constrain the signal amplitude within the linear operating range of the speaker, avoiding nonlinear distortion and hardware damage caused by power overload, and ensuring sound stability and hardware reliability. The limited time-domain signal is then converted from a discrete digital signal to a continuous analog voltage signal, matching the input format of the drive circuit and providing a compliant electrical signal foundation for subsequent hardware driving. Inputting the analog drive signal into the drive circuit to drive the tongue speaker completes the conversion from electrical signal to sound wave, allowing sound to radiate from the mouth through the oral cavity, achieving a human-like sound production position close to that of a real mouth.

[0105] Example 3

[0106] Figure 3 This is a schematic diagram of a robot-like humanoid speech generation device provided in Embodiment 3 of the present invention. Figure 3 As shown, the device includes: an acoustic feature sequence generation module 310, used to acquire input data and generate an acoustic feature sequence based on the input data;

[0107] The initial signal synthesis module 320 is used to synthesize an initial anthropomorphic speech signal based on acoustic feature sequences;

[0108] The voice signal compensation module 330 is used to compensate the frequency response and distortion of the initial anthropomorphic voice signal according to the preset acoustic compensation parameters of the robot's tongue speaker, and generate a compensated voice signal.

[0109] The voice signal playback module 340 is used to output the compensated voice signal to the tongue speaker for playback.

[0110] Optionally, the input data includes text data. The acoustic feature sequence generation module 310 specifically includes: a text data processing unit, used to: perform text processing and semantic recognition on the text data to generate linguistic features; determine the prosodic parameters and frame-level acoustic parameters corresponding to the linguistic features through a preset prosodic template or prosodic prediction model; and align the linguistic features, prosodic parameters, and frame-level acoustic parameters in a temporal sequence to generate an acoustic feature sequence.

[0111] Optionally, the input data includes reference speech. The acoustic feature sequence generation module 310 specifically includes: a reference speech processing unit, used for: performing frame segmentation processing on the reference speech to distinguish between spoken and silent frames; extracting fundamental frequency values ​​and frame-level acoustic parameters from the spoken frames; performing phoneme recognition on the reference speech to obtain linguistic features, and generating prosodic parameters and linguistic features by combining the fundamental frequency values; and temporally aligning the linguistic features, prosodic parameters, and frame-level acoustic parameters to generate an acoustic feature sequence.

[0112] Optionally, the acoustic feature sequence consists of time-aligned linguistic features, prosodic parameters, and frame-level acoustic parameters. The linguistic features include phonemes and tones, the prosodic parameters include duration, stress level, pauses, fundamental frequency trajectory, and amplitude envelope, and the frame-level acoustic parameters include spectral envelope, periodic energy ratio, and voice style parameters.

[0113] Optionally, the initial signal synthesis module 320 is specifically used to: generate a mixed speech excitation signal based on the amplitude envelope, fundamental frequency trajectory, and periodic energy ratio, wherein the mixed speech excitation includes voiced excitation and unvoiced excitation; construct duct filtering parameters based on the spectral envelope, perform duct filtering processing on the mixed speech excitation signal through the duct filtering parameters to obtain a filtered speech signal; and use the filtered speech signal as the initial anthropomorphic speech signal.

[0114] Optionally, the speech signal compensation module 330 is specifically used for: acquiring pre-calibrated acoustic compensation parameters, wherein the acoustic compensation parameters are constructed based on the actual acoustic path frequency response formed by the tongue speaker, tongue material, and robotic oral cavity; transforming the initial anthropomorphic speech signal to the frequency domain to obtain the initial speech spectrum; equalizing, gaining, and adjusting the dynamic range of the initial speech spectrum according to the acoustic compensation parameters to obtain the compensated speech spectrum; and inversely transforming the compensated speech spectrum into a time domain signal to generate the compensated speech signal.

[0115] Optionally, the voice signal playback module 340 is specifically used to: perform safe amplitude limiting processing on the compensated voice signal based on the preset amplitude range corresponding to the rated power of the tongue speaker, to obtain the amplitude-limited time-domain signal; perform digital-to-analog conversion on the amplitude-limited time-domain signal to obtain an analog drive signal; and input the analog drive signal into the drive circuit of the tongue speaker to drive the tongue speaker to complete the voice playback.

[0116] The technical solution of this invention enables multi-dimensional parameterized control by generating acoustic feature sequences, fully preserving the pronunciation content, rhythm, and timbre style, making the voice closer to the natural sounds of the human voice. By synthesizing initial anthropomorphic speech, simulating the mechanism of human vocalization, and integrating rhythm and breath characteristics, the mechanical feel of the speech is effectively reduced. Compensation reduces sound distortion caused by tongue material and oral cavity acoustic pathways, ensuring high sound quality reproduction. Finally, the sound is played through a tongue speaker, realizing mouth-based sound source production, thus improving the anthropomorphism and naturalness of the robot's speech.

[0117] The robot-like human voice generation device provided in this embodiment of the invention can execute the robot-like human voice generation method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0118] Example 4

[0119] Figure 4 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0120] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) or random access memory (RAM), communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from the storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. Input / output (I / O) interfaces are also connected to the bus 14.

[0121] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0122] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a method for generating humanoid speech in a robot.

[0123] In some embodiments, a method for generating humanoid speech in a robot can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the humanoid speech generation method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform a humanoid speech generation method in any other suitable manner (e.g., by means of firmware).

[0124] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0125] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0126] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0127] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0128] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0129] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system. It addresses the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0130] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0131] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for generating human-like speech in robots, characterized in that, Applications include robots equipped with tongue-shaped speakers, including: Acquire input data and generate an acoustic feature sequence based on the input data; An initial anthropomorphic speech signal is synthesized based on the acoustic feature sequence; The initial anthropomorphic speech signal is compensated to generate a compensated speech signal; The compensated speech signal is output to the tongue speaker for playback.

2. The method according to claim 1, characterized in that, The input data includes text data, and the step of generating an acoustic feature sequence based on the input data includes: The text data is processed and semantically recognized to generate linguistic features; The prosodic parameters and frame-level acoustic parameters corresponding to the linguistic features are determined by using a preset prosodic template or prosodic prediction model. The linguistic features, prosodic parameters, and frame-level acoustic parameters are temporally aligned to generate an acoustic feature sequence.

3. The method according to claim 1, characterized in that, The input data includes reference speech, and the step of generating an acoustic feature sequence based on the input data includes: The reference speech is segmented into frames to distinguish between frames with sound and frames without sound. Extract the fundamental frequency value and frame-level acoustic parameters from the audio frames; Phoneme recognition is performed on the reference speech to obtain linguistic features, and prosodic parameters and linguistic features are generated by combining the fundamental frequency value; The linguistic features, prosodic parameters, and frame-level acoustic parameters are temporally aligned to generate an acoustic feature sequence.

4. The method according to claim 2 or 3, characterized in that, The acoustic feature sequence consists of time-aligned linguistic features, prosodic parameters, and frame-level acoustic parameters. The linguistic features include phonemes and tones, the prosodic parameters include duration, stress level, pauses, fundamental frequency trajectory, and amplitude envelope, and the frame-level acoustic parameters include spectral envelope, periodic energy ratio, and voice style parameters.

5. The method according to claim 4, characterized in that, The synthesis of the initial anthropomorphic speech signal based on the acoustic feature sequence includes: Based on the amplitude envelope, fundamental frequency trajectory, and periodic energy ratio, a mixed speech excitation signal is generated, wherein the mixed speech excitation includes voiced excitation and unvoiced excitation; Based on the spectral envelope, vocal tract filtering parameters are constructed, and the mixed speech excitation signal is subjected to vocal tract filtering processing using the vocal tract filtering parameters to obtain the filtered speech signal. The filtered speech signal is used as the initial anthropomorphic speech signal.

6. The method according to claim 1, characterized in that, The step of compensating the initial anthropomorphic speech signal to generate a compensated speech signal includes: Obtain pre-calibrated acoustic compensation parameters, wherein the acoustic compensation parameters are constructed based on the actual acoustic path frequency response formed by the tongue speaker, tongue material and robotic oral cavity; The initial anthropomorphic speech signal is transformed to the frequency domain to obtain the initial speech spectrum; The initial speech spectrum is equalized, and its gain and dynamic range are adjusted according to the acoustic compensation parameters to obtain the compensated speech spectrum. The compensated speech spectrum is inversely transformed into a time-domain signal to generate the compensated speech signal.

7. The method according to claim 1, characterized in that, The step of outputting the compensated speech signal to the tongue speaker for playback includes: Based on the preset amplitude range corresponding to the rated power of the tongue speaker, the compensated speech signal is subjected to safe amplitude limiting processing to obtain the amplitude-limited time domain signal. The time-domain signal after clipping is converted from digital to analog to obtain an analog drive signal; The analog drive signal is input into the drive circuit of the tongue speaker to drive the tongue speaker to complete the voice playback.

8. A robot-inspired human voice generation device, characterized in that, include: An acoustic feature sequence generation module is used to acquire input data and generate an acoustic feature sequence based on the input data; An initial signal synthesis module is used to synthesize an initial anthropomorphic speech signal based on the acoustic feature sequence; The voice signal compensation module is used to compensate the initial anthropomorphic voice signal for frequency response and distortion based on the preset acoustic compensation parameters of the robot's tongue speaker, and generate a compensated voice signal. The voice signal playback module is used to output the compensated voice signal to the tongue speaker for playback.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer instructions that are used to cause a processor to execute the method of any one of claims 1-7.

Images