Voiceprint registration method, device, equipment, storage medium and program product

Abstract

The present disclosure provides a voiceprint registration method, device, equipment, storage medium and program product, relating to the technical field of computer. The method comprises: in response to receiving a voiceprint registration instruction of a user, receiving voice information; dividing the voice information into multiple voice segments; in response to the average environmental noise level of the multiple voice segments not exceeding a first threshold value, and any continuous voice segment corresponding to the same user, performing voiceprint registration based on the voice information. The method can effectively improve the voice quality of voiceprint registration.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a voiceprint registration method, apparatus, device, storage medium, and program product. Background Technology

[0002] With the popularization of voice interaction technology, voiceprint recognition has been widely used in smart terminals, in-vehicle systems, financial authentication, and other scenarios. In a voiceprint recognition system, users must first complete voiceprint registration to establish their personal voiceprint model. The voice quality during the registration phase directly determines the performance of subsequent recognition or noise reduction.

[0003] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this disclosure is to provide a voiceprint registration method, apparatus, device, storage medium, and program product.

[0005] According to a first aspect of the present disclosure, a voiceprint registration method is provided, comprising: receiving voice information in response to receiving a user's voiceprint registration instruction; dividing the voice information into multiple voice segments; and performing voiceprint registration based on the voice information in response to the fact that the environmental noise levels of the multiple voice segments do not exceed a first threshold and any consecutive voice segments correspond to the same user.

[0006] In some implementations, in response to the fact that the environmental noise levels of multiple speech segments do not exceed a first threshold and any consecutive speech segments correspond to the same user, voiceprint registration is performed based on the speech information, including: in response to the fact that the environmental noise levels of multiple speech segments do not exceed a first threshold and any consecutive speech segments correspond to the same user, calculating the total duration of multiple speech segments; and in response to the total duration reaching a preset duration threshold, performing voiceprint registration based on multiple speech segments.

[0007] In some implementations, the method further includes: converting the current speech segment into frequency domain data; performing noise reduction processing on the frequency domain data to obtain noise-reduced frequency domain data; converting the noise-reduced frequency domain data back to the time domain to obtain noise-reduced speech; calculating the original energy of the current speech segment and the energy of the noise-reduced speech respectively; dividing the difference between the original energy and the energy of the noise-reduced speech by the original energy to obtain a noise energy ratio; and determining that the environmental noise level of the current speech segment exceeds the first threshold when the noise energy ratio is greater than a first threshold.

[0008] In some implementations, the method further includes: extracting the voiceprint features of the current speech segment using a voiceprint feature extraction model, and extracting the voiceprint features of the previous speech segment using the same model; calculating the similarity between the voiceprint features of the current speech segment and the voiceprint features of the previous speech segment; and determining that the current speech segment and the previous speech segment originate from the same user when the similarity is not lower than a second threshold.

[0009] In some implementations, the voiceprint feature extraction model is used to extract the voiceprint features of the current speech segment, including: performing a short-time Fourier transform on the current speech segment to obtain the frequency domain data of the current speech segment; calculating the power spectrum based on the frequency domain data, and extracting filter bank features based on the power spectrum; and inputting the filter bank features into the voiceprint feature extraction model to obtain the voiceprint features of the current speech segment.

[0010] In some implementations, dividing speech information into multiple speech segments includes dividing continuous speech in the speech information into multiple speech segments of fixed duration.

[0011] In some implementations, the method further includes: in response to the ambient noise level of any voice segment exceeding the first threshold, or any consecutive voice segments corresponding to different users, outputting a prompt message to the user, the prompt message indicating that the registration failure was due to excessive ambient noise or the detection of non-target user voice.

[0012] According to a second aspect of the present disclosure, a voiceprint registration device is provided, comprising: a voice receiving module, a voice processing module, and a voiceprint registration module.

[0013] The voice receiving module is used to receive voice information in response to receiving a user's voiceprint registration command; The speech processing module is used to divide speech information into multiple speech segments; The voiceprint registration module is used to register voiceprints based on voice information when the environmental noise levels of multiple voice segments do not exceed a first threshold and any consecutive voice segments correspond to the same user.

[0014] In some implementations, the voiceprint registration module is used to calculate the total duration of multiple voice segments in response to the fact that the environmental noise levels of multiple voice segments do not exceed a first threshold and any consecutive voice segments correspond to the same user; and to perform voiceprint registration based on multiple voice segments in response to the total duration reaching a preset duration threshold.

[0015] In some embodiments, the voiceprint registration device further includes: The noise detection module is used to convert the current speech segment into frequency domain data; perform noise reduction processing on the frequency domain data to obtain denoised frequency domain data; convert the denoised frequency domain data back to the time domain to obtain denoised speech; calculate the original energy of the current speech segment and the energy of the denoised speech respectively; divide the difference between the original energy and the energy of the denoised speech by the original energy to obtain the noise energy ratio; when the noise energy ratio is greater than a first threshold, it is determined that the environmental noise level of the current speech segment exceeds the first threshold. In some embodiments, the voiceprint registration device further includes: The user judgment module is used to extract the voiceprint features of the current speech segment using a voiceprint feature extraction model, and to extract the voiceprint features of the previous speech segment using the same model; calculate the similarity between the voiceprint features of the current speech segment and the voiceprint features of the previous speech segment; and determine that the current speech segment and the previous speech segment originate from the same user when the similarity is not lower than a second threshold.

[0016] In some implementations, the user judgment module uses a voiceprint feature extraction model to extract the voiceprint features of the current speech segment, including: performing a short-time Fourier transform on the current speech segment to obtain the frequency domain data of the current speech segment; calculating the power spectrum based on the frequency domain data, and extracting filter bank features based on the power spectrum; and inputting the filter bank features into the voiceprint feature extraction model to obtain the voiceprint features of the current speech segment.

[0017] In some implementations, the speech processing module is used to divide continuous speech in the speech information into multiple speech segments of fixed duration.

[0018] In some embodiments, the voiceprint registration device further includes: a prompt output module, configured to output a prompt message to the user in response to any ambient noise level of any voice segment exceeding the first threshold, or any consecutive voice segments corresponding to different users, the prompt message indicating that the registration failure is due to excessive ambient noise or the detection of non-target user voice.

[0019] According to a third aspect of the present disclosure, a vehicle is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above-described voiceprint registration method.

[0020] According to a fourth aspect of the present disclosure, an electronic device is provided, characterized in that it includes: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above-described voiceprint registration method.

[0021] According to a fifth aspect of the present disclosure, a computer-readable storage medium is provided, which, when the instructions in the storage medium are executed by a processor of a mobile terminal, enables the mobile terminal to perform the above-described voiceprint registration method.

[0022] According to a sixth aspect of the present disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the above-described voiceprint registration method.

[0023] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects: This disclosure divides the received speech information into multiple speech segments and detects whether the environmental noise level of each segment exceeds a first threshold, thus selecting only low-noise segments for registration and effectively eliminating background noise interference. By verifying whether any consecutive speech segments correspond to the same user, it ensures the consistency of the speaker (i.e., user) in the registration data and prevents model contamination caused by the mixing of non-user voices. In addition, through the dual screening mechanism of "noise filtering" and "voiceprint consistency verification" mentioned above, this disclosure not only improves the success rate and accuracy of voiceprint registration, but also enhances robustness in complex acoustic environments, laying a solid foundation for subsequent high-precision voiceprint recognition.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0026] Figure 1 This is a flowchart illustrating a voiceprint registration method according to some embodiments of the present disclosure.

[0027] Figure 2 This is a flowchart illustrating a noise estimation judgment based on some embodiments of the present disclosure.

[0028] Figure 3 This is a flowchart illustrating a model training process according to some embodiments of the present disclosure.

[0029] Figure 4 This is a flowchart illustrating a voiceprint consistency verification according to some embodiments of the present disclosure.

[0030] Figure 5 This is a flowchart illustrating a voiceprint feature extraction method according to some embodiments of the present disclosure.

[0031] Figure 6 This is a flowchart illustrating another noise estimation judgment based on some embodiments of the present disclosure.

[0032] Figure 7 This is a flowchart illustrating a voiceprint information determination method according to some embodiments of the present disclosure.

[0033] Figure 8 This is a block diagram illustrating a voiceprint registration device according to some embodiments of the present disclosure.

[0034] Figure 9 This is a block diagram illustrating an electronic device according to some embodiments of the present disclosure. Detailed Implementation

[0035] Exemplary embodiments of this disclosure will be described in detail herein, examples of which are illustrated in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings denote the same or similar elements unless otherwise indicated. Various changes, modifications, and equivalents of the methods, apparatus, and / or systems described herein will become apparent upon understanding this disclosure. For example, the order of operations described herein is merely illustrative and is not limited to those orders set forth herein, but can be changed as will become apparent upon understanding this disclosure, except for operations that must be performed in a particular order. Furthermore, for clarity and brevity, descriptions of features known in the art may be omitted.

[0036] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all content and steps, nor does it necessarily have to be executed in the described order or in the order of the step numbers. For example, some steps can be broken down, while others can be combined or partially combined, and multiple steps can have their order interchanged or be executed simultaneously. Therefore, the actual execution order may change depending on the actual situation.

[0037] The embodiments described below, which are examples of some of the embodiments of this disclosure, do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0038] In applications such as voiceprint recognition and voiceprint denoising, users typically need to pre-register their voiceprints to build their personal voiceprint model. However, voiceprint registration places high demands on the acquisition environment: ideally, the registered speech should only contain clear speech from the target speaker (i.e., the user). The inventors discovered that in real-world usage environments, users often register under non-ideal conditions, such as background noise, multiple speakers simultaneously, or the recording device being too far away. If registration is performed in scenarios with ambient noise or multiple speakers, the acquired speech will be mixed with non-target acoustic components, causing the constructed voiceprint model to deviate from the user's true speech characteristics, thus significantly reducing the performance of subsequent voiceprint recognition or voiceprint denoising.

[0039] Traditional acoustic environment detection methods typically rely on noise estimation techniques, analyzing noise energy or signal-to-noise ratio in speech signals to determine the quietness of the environment. However, these methods only reflect the ambient noise level and cannot identify the presence of other speakers interfering. In voiceprint registration scenarios, even with low ambient noise, the simultaneous speech of others can still contaminate the registration data. Therefore, relying solely on noise estimation is insufficient to meet the dual environmental conditions of "single speaker and low noise" required for voiceprint registration.

[0040] Furthermore, ordinary users often struggle to accurately determine whether their current environment meets the registration requirements when registering their voiceprints. Without an effective environmental compliance detection mechanism, the system may allow registration to complete in any environment, resulting in a voiceprint model containing interfering information. For example, voiceprint features obtained from registration in noisy environments or multi-person conversation scenarios actually mix background noise, other people's speech, and other non-target components, lacking the ability to purely represent the target user's speech. This ultimately leads to a significant decrease in the effectiveness of downstream tasks such as voiceprint denoising.

[0041] To address the aforementioned issues, the solution provided in this disclosure can automatically detect the presence of multiple people speaking in the surrounding environment and estimate noise; it also provides appropriate environmental detection and judgment for the surrounding environment when registering the target speaker for voiceprint noise reduction.

[0042] To address the aforementioned issues, the solution provided in this disclosure can automatically assess the noise level of the surrounding environment during the registration process and detect the presence of speech from non-target speakers, thereby determining whether the current environment meets the conditions for voiceprint registration. This helps ensure that the collected registration speech only contains clear speech from the target speaker, improving the purity and reliability of subsequent voiceprint models.

[0043] In some embodiments of this disclosure, the acquisition of data and information, as well as the collection, updating, analysis, processing, use, transmission, and storage of related user personal information, may comply with the laws and regulations of the country where the location is situated.

[0044] In some embodiments of this disclosure, data, information, etc., may be obtained after obtaining the user's consent.

[0045] In some embodiments of this disclosure, the solutions of this disclosure can be applied to fields such as mobile phones, tablets, acoustics, and signal processing. Specifically, they can be applied to environmental detection of devices such as speakers, as well as the identification or rejection of device scenarios such as mobile phones, speakers, and vehicle wake-up.

[0046] The specific implementation methods of the embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0047] Figure 1This is a flowchart illustrating a voiceprint registration method according to some embodiments of the present disclosure, such as... Figure 1 As shown, this voiceprint registration method can be applied to electronic devices, including but not limited to in-vehicle terminals, smartphones, smart tablets, wearable devices, desktop computers, laptops, smart speakers, and other terminal devices. It can also include server-side components such as local servers and cloud servers, which can be deployed in a computer cluster or a combination of multiple computers. The voiceprint registration method may include the following steps S110-S130.

[0048] In S110, in response to receiving the user's voiceprint registration instruction, voice information is received.

[0049] In some embodiments, users can trigger a voiceprint registration command through a front-end interface (such as an App button or a voice assistant wake-up word), and then the terminal device can call the microphone array to start the audio stream acquisition service, that is, receive voice information.

[0050] In S120, the speech information is divided into multiple speech segments.

[0051] The aforementioned voice information can be continuous voice input by the user, and this step can be to divide the continuous voice into multiple voice segments.

[0052] In some embodiments, continuous speech is emitted by the user in real time during the voiceprint registration process and picked up by the device’s microphone array or audio acquisition module; the segmentation operation can be based on a time window to form several discrete speech processing units, which facilitates subsequent segment-by-segment assessment of environmental compliance.

[0053] In an exemplary embodiment, continuous speech is segmented into segments of fixed duration (e.g., 2 to 5 seconds), and each segment participates in the subsequent judgment process as an independent analysis object.

[0054] In S130, in response to the fact that the environmental noise levels of multiple voice segments do not exceed the first threshold and any consecutive voice segments correspond to the same user, voiceprint registration is performed based on the voice information.

[0055] In this embodiment of the disclosure, before voiceprint registration, it is first determined whether the ambient noise level of the current voice segment exceeds a first threshold, and whether the current voice segment and the previous voice segment originate from the same user. Only when the ambient noise levels of multiple voice segments do not exceed the first threshold, and any consecutive voice segments correspond to the same user, is it determined that the current voice information can be used for voiceprint registration, and the voiceprint registration process continues.

[0056] In some embodiments, the ambient noise level reflects the degree of interference from non-target user components (such as background noise and ambient noise) in the current speech segment; by setting a first threshold as the upper limit of noise tolerance, speech segments with high signal-to-noise ratio can be effectively screened out for registration.

[0057] In an exemplary embodiment, the current speech segment is converted into frequency domain data and then input into a pre-trained noise reduction neural network to obtain the noise-reduced speech. The ratio of the original energy to the noise-reduced energy is then calculated as the noise energy ratio. When the ratio is greater than a first threshold, the environmental noise level is determined to be excessive.

[0058] In some embodiments, it is determined whether the current speech segment and the previous speech segment originate from the same user. This is used to verify the consistency of the identity of the speech source during the registration process and to prevent the mixing of other people's speech into the voiceprint model. The core of the above determination lies in comparing the acoustic feature similarity of adjacent speech segments.

[0059] In an exemplary embodiment, a voiceprint feature extraction model (such as ECAPA-TDNN) is used to extract the voiceprint feature vectors of the current speech segment and the previous speech segment respectively, and the cosine similarity between the two is calculated; when the similarity is not lower than the second threshold, it is determined that the two originate from the same user.

[0060] The environmental noise level of the current speech segment does not exceed the first threshold, including when the environmental noise level of the current speech segment is less than or equal to the first threshold.

[0061] In some embodiments, the aforementioned dual conditions constitute a joint criterion for registration admission, ensuring that the adopted voice segments simultaneously meet the two key quality requirements of "low noise" and "single user".

[0062] In an exemplary embodiment, only speech segments that pass through both the noise judgment module and the user judgment module are accumulated into the effective registration duration and used for the construction of the final voiceprint model.

[0063] In some embodiments, if the ambient noise level of any voice segment exceeds a first threshold, or if the current voice segment and the previous voice segment do not originate from the same user, the voiceprint registration process is terminated. A "zero-tolerance" strategy is adopted, and once any violation is detected, the entire registration process is immediately interrupted to avoid the accumulation of contaminated data that could affect the reliability of the model.

[0064] In an exemplary embodiment, after an anomaly is detected, the reason for the failure (such as "excessive noise" or "non-target user detected") can be recorded, and a user prompt mechanism can be triggered.

[0065] As can be seen from the above steps, the voiceprint registration method provided in this disclosure can automatically perform fine-grained quality control on the registered speech: on the one hand, it eliminates the influence of high-noise environments, and on the other hand, it prevents the mixing of non-target user speech, thereby ensuring that the constructed voiceprint model is based only on the pure speech of the target user, which significantly improves the accuracy and robustness of subsequent voiceprint recognition, voiceprint noise reduction and other applications.

[0066] In some embodiments, the above-mentioned response to the environmental noise levels of multiple speech segments not exceeding a first threshold, and any consecutive speech segments corresponding to the same user, for voiceprint registration based on speech information, may include: calculating the total duration of the multiple speech segments in response to the environmental noise levels of multiple speech segments not exceeding the first threshold, and any consecutive speech segments corresponding to the same user; and performing voiceprint registration based on the multiple speech segments in response to the total duration reaching a preset duration threshold. The preset duration threshold may be the minimum speech duration required for voiceprint registration.

[0067] The embodiments of this disclosure first perform strict preprocessing and screening on the received voice information. Only voice segments that simultaneously meet the conditions of low noise and user consistency will be recognized as valid data and included in the subsequent processing flow (i.e., the judgment of whether the valid duration is sufficient), thereby ensuring the quality of registration data from the source.

[0068] To address the issue of insufficient effective voice duration, a flexible response mechanism is introduced. When the calculated total duration of effective voice segments is less than the minimum voice duration threshold required for voiceprint registration, it is determined that the current data volume is insufficient to build a stable voiceprint model. In this case, an error message can be displayed, prompting the user to re-enter complete voice information; alternatively, the receiving state can be maintained, guiding the user to continue speaking or wait for subsequent voice input. In this mode, the duration of newly acquired voice segments that meet quality requirements can be accumulated until the total effective duration reaches the preset threshold before triggering the final registration operation, avoiding the drawback of having to start the entire registration process from scratch due to a single recording being too short.

[0069] For situations with ample effective voice duration, an immediate truncation strategy is employed to improve efficiency. When the total accumulated effective voice segment duration exceeds or equals the minimum voice duration threshold, there's no need to wait for the user to finish speaking all preset content or process all redundant data. Once it's confirmed that the accumulated duration just meets the threshold requirement, this high-quality data can be immediately truncated to initiate the voiceprint registration process. This mechanism ensures registration is completed as soon as the data volume meets the requirement, shortening user waiting time and reducing the system's overhead in processing excess voice data.

[0070] In some embodiments, the cumulative duration includes only qualified speech segments that have passed dual verification. The minimum speech duration required for registration is the minimum amount of effective speech required to build a stable speakerprint model. As an example, the minimum speech duration required for registration could be a value between 10 and 20 seconds.

[0071] In an exemplary embodiment, an effective voice timer can be maintained in real time. Whenever a voice segment passes the judgment, its duration is accumulated. When the accumulated value is greater than or equal to the minimum voice duration (e.g., 15 seconds), the voiceprint model is triggered and the registration process ends.

[0072] The above solution enables dynamic adaptive management of speech duration. Regardless of whether the user inputs short, segmented sentences or continuous long sentences, the system can intelligently determine whether to continue accumulating data or immediately execute registration. This not only ensures the sufficiency and stability of the voiceprint model training data but also significantly improves the registration success rate and user experience smoothness across various speaking habits and environments.

[0073] In some embodiments, determining whether the ambient noise level of the current speech segment exceeds a first threshold may include... Figure 2 S201-S206 are shown.

[0074] In S201, the current speech segment is converted into frequency domain data.

[0075] In some embodiments, the transformation operation is used to map the time-domain speech signal to a frequency-domain representation so that the subsequent noise reduction neural network can perform spectral-level noise suppression processing; the frequency-domain data may include information in the form of amplitude spectrum, power spectrum or complex spectrum.

[0076] In an exemplary embodiment, the current speech segment is processed using a Short-Time Fourier Transform (STFT) with a frame length of 25 milliseconds and a frame shift of 10 milliseconds to obtain a complex frequency domain data matrix.

[0077] In S202, the frequency domain data is denoised to obtain the denoised frequency domain data.

[0078] In some embodiments, the above-mentioned noise reduction process may involve inputting frequency domain data into a pre-trained noise reduction neural network to obtain noise-reduced frequency domain data.

[0079] In some embodiments, the denoising neural network is a deep learning model trained on a large number of noisy-clean speech pairs, with the goal of recovering a near-clean frequency domain representation of the target user's speech from noisy input.

[0080] In an exemplary embodiment, the denoising neural network is a convolutional neural network (CNN), whose input is the STFT coefficients of the noisy speech and whose output is the estimated STFT coefficients of the clean speech; the network has been trained end-to-end on a simulated noisy speech dataset.

[0081] In S203, the denoised frequency domain data is converted back to the time domain to obtain the denoised speech.

[0082] In some embodiments, the transformation operation is used to reconstruct the denoised time-domain waveform so that it can be compared with the original speech in the same domain, thereby quantifying the proportion of noise components.

[0083] In an exemplary embodiment, the inverse short-time Fourier transform (iSTFT) is used to reconstruct the denoised frequency domain data, and combined with the overlap-add method, a continuous denoised time-domain speech signal is generated.

[0084] In S204, the original energy of the current speech segment and the energy of the denoised speech segment are calculated respectively.

[0085] In some embodiments, energy refers to the short-time energy of the speech signal, typically defined as the sum of squares of the amplitudes at sampling points, and is a fundamental acoustic feature for measuring speech activity and the degree of noise interference.

[0086] In some embodiments, the energy of speech refers to the overall activity of the speech signal in the time domain, which is usually obtained by squaring and summing the amplitudes of the speech sampling points; the original energy reflects the total acoustic energy including noise, while the energy of the denoised speech mainly reflects the pure energy of the target user's speech.

[0087] In an exemplary embodiment, the original energy can be obtained by summing the squared amplitudes of all sampling points of the current speech segment; similarly, the same operation can be performed on the reconstructed time-domain speech after noise reduction to obtain the noise-reduced energy.

[0088] In S205, the difference between the original energy and the energy of the denoised speech is divided by the original energy to obtain the noise energy ratio.

[0089] In some embodiments, this ratio is used to measure the proportion of energy in the original speech that is identified as noise, and is a normalized noise metric. Because it is based on the relative relationship between the energy difference and the original energy, it can effectively eliminate the influence of different user speaking volumes, making noise judgment more robust.

[0090] In an exemplary embodiment, the difference between the original speech energy and the noise-reduced speech energy can be calculated first, and then the difference can be divided by the original speech energy to obtain a value between 0 and 1. This value is the noise energy ratio. The larger the value, the more severe the noise interference in the environment.

[0091] In S206, when the noise energy ratio is greater than the first threshold, it is determined that the environmental noise level of the current speech segment exceeds the first threshold.

[0092] In some embodiments, the first threshold is a preset empirical value or a configurable parameter used to define an acceptable upper limit for noise; once the ratio exceeds the threshold, the current environment is considered unsuitable for voiceprint registration.

[0093] As can be seen from the above steps, this disclosure, by introducing an energy ratio analysis method based on a noise reduction neural network, can more accurately quantify the degree of interference in complex noise scenarios. Compared with traditional methods that rely solely on signal-to-noise ratio or zero-crossing rate, it significantly improves the robustness and adaptability of noise judgment, thereby providing reliable environmental quality assurance for voiceprint registration.

[0094] In some embodiments, the pre-trained denoising neural network is a convolutional neural network; the training process of the denoising neural network may include... Figure 3 S301-S305 are shown.

[0095] In S301, obtain clean speech samples.

[0096] In some embodiments, a clean speech sample refers to high-quality speech data that is free from background noise, interference from other people's speech, and contains only the clear pronunciation of a single user, and is used as an ideal target signal for model training.

[0097] In an exemplary embodiment, clean speech samples may be derived from publicly available speech databases or internally collected controlled environment recordings, covering users of different genders, ages, accents, and speaking speeds, in order to improve the model's generalization ability.

[0098] In S302, clean speech samples are mixed with noisy samples to simulate the generation of noisy speech.

[0099] In some embodiments, by superimposing clean speech with real or synthetic noise in a certain proportion, a large-scale noisy-clean speech pair can be constructed for supervised training of a noise reduction model; noise samples include, but are not limited to, interference sources in typical application scenarios such as street noise, restaurant noise, keyboard typing, and background noise from multiple conversations.

[0100] In an exemplary embodiment, a noise sample can be randomly selected, its energy level adjusted, and then weighted and superimposed with a clean speech sample to enable the synthesized noisy speech to cover a variety of actual noise intensities.

[0101] In S303, a Short-Time Fourier Transform (STFT) is performed on the noisy speech to obtain the frequency domain data of the noisy speech.

[0102] In this embodiment of the disclosure, STFT is used to convert noisy speech in the time domain into a time-frequency representation, which facilitates the neural network to identify and suppress noise components in the spectral dimension; the resulting frequency domain data usually contains amplitude information, which can be used as input features of the noise reduction network.

[0103] As an example, a Hamming window can be used to process noisy speech in frames, and then STFT can be performed on each frame to form a time-frequency matrix, which can be used as the input of a convolutional neural network.

[0104] In S304, a short-time Fourier transform is performed on the clean speech sample to obtain the frequency domain data of the clean speech.

[0105] In some embodiments, the same STFT processing is performed on clean speech in order to obtain a target output that is aligned with the time-frequency structure of the input noisy speech, thereby guiding the network to learn the mapping relationship from the noisy spectrum to the clean spectrum.

[0106] In an exemplary embodiment, the STFT parameters (including window length, frame shift, and FFT points) of the clean speech are completely consistent with those of the noisy speech, ensuring that the input and the target correspond strictly in the time and frequency dimensions.

[0107] In S305, the convolutional neural network is trained using the frequency domain data of noisy speech as input and the frequency domain data of clean speech as training target to obtain the denoising neural network.

[0108] In some embodiments, the above training process employs supervised learning, continuously optimizing network parameters by minimizing the difference (such as mean square error) between the network output and clean speech domain data, enabling it to estimate a spectrum close to the original clean speech from any noisy speech.

[0109] In an exemplary embodiment, the convolutional neural network used includes multiple layers of one-dimensional or two-dimensional convolutional modules and has a channel attention mechanism, which can adaptively enhance the target speech and suppress noise in the frequency band and time dimensions. After training, the network serves as the "pre-trained noise reduction neural network" in the above embodiment and is used for noise level evaluation in the voiceprint registration process.

[0110] As can be seen from the above steps, this disclosure obtains a noise reduction model that can effectively separate target speech from noise by training a convolutional neural network end-to-end on simulated noisy data. This model can not only be used for speech enhancement, but also as a noise perception tool to provide a reliable environmental quality judgment basis for voiceprint registration, thereby improving the robustness and accuracy of the entire registration process.

[0111] In some embodiments, determining whether the current voice segment and the previous voice segment originate from the same user may include... Figure 4 S401-S403 are shown.

[0112] In S401, the voiceprint feature extraction model is used to extract the voiceprint features of the current speech segment, and the voiceprint feature extraction model is used to extract the voiceprint features of the previous speech segment.

[0113] In some embodiments, voiceprint features refer to acoustic embedding vectors that can characterize user identity and have high discriminativeness and robustness. By extracting features from two speech segments separately using the same model, the consistency of the feature space can be ensured, providing a reliable foundation for subsequent similarity calculations.

[0114] In an exemplary embodiment, the voiceprint feature extraction model is a deep neural network model; the model receives preprocessed speech features (such as filter bank features) as input and outputs a fixed-dimensional voiceprint embedding vector (e.g., 192-dimensional) to represent the user identity information of the corresponding speech segment.

[0115] In S402, the similarity between the voiceprint features of the current speech segment and the voiceprint features of the previous speech segment is calculated.

[0116] In some embodiments, similarity is used to quantify how close two voiceprint features are in the feature space; a higher value indicates that they are more likely to come from the same user. This calculation process does not depend on specific user identity labels, but is based solely on the geometric relationships of the features themselves.

[0117] In an exemplary embodiment, cosine similarity is used as a similarity metric, which is to calculate the cosine value of the angle between two voiceprint embedding vectors; this value is normalized to the range of [-1, 1], and the closer the value is to 1, the more similar the voiceprint features are.

[0118] In S403, when the similarity is not lower than the second threshold, it is determined that the current speech segment and the previous speech segment come from the same user.

[0119] In some embodiments, the second threshold is a pre-set lower limit of similarity, used to distinguish the boundary between "the same user" and "different users"; the threshold can be adjusted according to the security requirements or the tolerance of false recognition rate of the application scenario.

[0120] If the calculated similarity is greater than or equal to the second threshold, it is determined that the two voice segments originate from the same user, and the current voice segment is allowed to participate in voiceprint registration; otherwise, it is considered that a non-target user's voice has been detected, triggering the registration process termination mechanism.

[0121] As can be seen from the above steps, this disclosure, by introducing a cross-segment consistency verification mechanism based on deep voiceprint features, can dynamically monitor the continuity of user identity during the registration process, effectively prevent voiceprint model contamination in scenarios where others interrupt or multiple people are talking, thereby significantly improving the purity of registration data and the reliability of subsequent voiceprint applications.

[0122] In some embodiments, extracting the voiceprint features of the current speech segment using a voiceprint feature extraction model may include: Figure 5 S501-S503 are shown.

[0123] In S501, a short-time Fourier transform is performed on the current speech segment to obtain the frequency domain data of the current speech segment.

[0124] In some embodiments, the short-time Fourier transform is used to convert time-domain speech into frequency-domain data so that acoustic features with speech-aware characteristics can be extracted subsequently; the resulting frequency-domain data reflects the energy distribution of speech at different frequencies and times.

[0125] In an exemplary embodiment, the current speech segment is framed using a Hamming window, and a short-time Fourier transform is performed on each frame to generate a complex-form spectrum matrix, i.e., frequency domain data, which serves as the basis for subsequent power spectrum calculations.

[0126] In S502, the power spectrum is calculated based on the frequency domain data, and filter bank features (FBank features) are extracted based on the power spectrum.

[0127] In some embodiments, the power spectrum is obtained by squaring the amplitude of the frequency domain data, which can more accurately reflect the energy intensity of each frequency band; the filter bank characteristics are obtained by weighted integration and logarithmic calculation of the power spectrum through a Mel filter bank that simulates the hearing characteristics of the human ear, which has good user discrimination ability.

[0128] As an example, the power spectrum is obtained by first squaring the amplitude of the short-time Fourier transform result; then the power spectrum is filtered through a set of Mel filters, and the natural logarithm of the output of each filter is taken to finally form the log-Mel filter bank characteristic sequence.

[0129] In S503, the filter bank features are input into the voiceprint feature extraction model to obtain the voiceprint features of the current speech segment.

[0130] In some embodiments, the voiceprint feature extraction model is a deep neural network trained on large-scale user data, which can automatically learn high-order, robust user embedding representations from filter bank features; the output voiceprint features are fixed-dimensional vectors that can be used for similarity comparison or model registration.

[0131] In an exemplary embodiment, the voiceprint feature extraction model receives the filter bank features of the entire speech segment as input, and outputs a 192-dimensional voiceprint embedding vector through mechanisms such as time-delay convolution, channel attention, and multi-scale aggregation. This vector is the voiceprint feature of the current speech segment, which is used for subsequent consistency verification with the features of the previous speech segment.

[0132] As can be seen from the above steps, this disclosure converts the original speech into an acoustic representation that conforms to the characteristics of human hearing by sequentially executing the front-end processing flow of short-time Fourier transform, power spectrum calculation and filter bank feature extraction. This representation is then used as the input to the voiceprint feature extraction model, so that the generated voiceprint features can effectively reflect the identity characteristics of the target user, thereby supporting the accurate judgment of user consistency between adjacent speech segments.

[0133] In some embodiments, dividing speech information into multiple speech segments can be achieved by dividing continuous speech within the speech information into multiple speech segments of fixed duration. The fixed duration can be set according to the requirements of voiceprint registration for speech segment length and real-time processing capabilities. During segmentation, continuous speech streams can be divided using equal-length windows, forming speech units with uniform structure that are easy to process in parallel.

[0134] In an exemplary embodiment, the continuous speech input by the user is divided into sliding or non-overlapping segments with a fixed segment duration of 3 seconds. Each speech segment independently participates in the subsequent environmental noise judgment and user consistency verification process.

[0135] By dividing continuous speech into multiple speech segments with fixed durations, this disclosure simplifies the scheduling and processing logic of speech streams and avoids the boundary uncertainties caused by dynamic segmentation. At the same time, the uniform segment length facilitates the parameter configuration and efficient execution of subsequent modules such as noise estimation, voiceprint feature extraction, and similarity calculation, and ensures that each speech segment is comparable in the time dimension. This supports the accurate accumulation of effective registration duration and timely response to abnormal situations, ultimately improving the stability and reliability of the voiceprint registration process while ensuring user experience.

[0136] In some embodiments, the fixed segment duration is 2 to 5 seconds, used to divide the continuous voice input by the user into multiple independently processed voice segments; at the same time, the minimum effective voice cumulative duration required for voiceprint registration is set to 10 to 20 seconds.

[0137] The above parameter ranges are set based on a balance between the technical requirements of voiceprint modeling and practical usage scenarios. If the speech segment is shorter than 2 seconds, it may not be able to fully cover the user's pronunciation characteristics, resulting in unstable voiceprint feature extraction; if it is longer than 5 seconds, it will be difficult to respond promptly to sudden noise in the environment or interference from other people's voices. When the minimum registration time is less than 10 seconds, the collected speech often lacks sufficient phoneme diversity, affecting the model's discrimination ability; exceeding 20 seconds will significantly prolong the registration time and reduce the user experience, especially in interaction-restricted scenarios such as in-vehicle and mobile environments.

[0138] By combining segment durations of 2 to 5 seconds with a minimum registration duration of 10 to 20 seconds, fine-grained monitoring of environmental compliance can be achieved while ensuring the quality of the voiceprint model. Each voice segment can be independently verified and selectively accumulated, which supports both rapid termination of abnormal processes and efficient completion of registration within 3 to 5 qualified segments, taking into account model robustness, environmental adaptability and user operation convenience.

[0139] In some embodiments, multiple voice segments are sequentially arranged in chronological order as a first voice segment, a second voice segment, and subsequent voice segments. The first voice segment is not used to perform the step of determining whether the ambient noise level of the current voice segment exceeds a first threshold, nor is it used to perform the step of determining whether the current voice segment and the previous voice segment originate from the same user. The second voice segment is used to perform the step of determining whether the ambient noise level of the current voice segment exceeds the first threshold, but is not used to perform the step of determining whether the current voice segment and the previous voice segment originate from the same user. Starting from subsequent voice segments, each voice segment performs the steps of determining whether the ambient noise level of the current voice segment exceeds the first threshold and determining whether the current voice segment and the previous voice segment originate from the same user.

[0140] In this embodiment, the voice segments are divided into first, second, and subsequent voice segments in chronological order, and a phased activation detection mechanism is adopted: the first voice segment is skipped and does not participate in any judgment; the second voice segment only performs noise level assessment; from the third segment onwards, noise judgment and user consistency verification are performed simultaneously. This strategy avoids misjudgment caused by a lack of reference data in the early stages of registration (such as the inability to compare voiceprints if the first segment has no preceding segment), while ensuring that a complete environmental compliance verification is performed from a valid starting point, thus improving the stability and accuracy of the registration process.

[0141] In some embodiments, the above-described voiceprint registration method may further include: in response to the ambient noise level of any voice segment exceeding the first threshold, or any consecutive voice segments corresponding to different users, outputting a prompt message to the user, the prompt message indicating that the registration failure is due to excessive ambient noise or the detection of non-target user voice.

[0142] In this embodiment of the disclosure, when any voice segment is detected as not meeting the registration conditions, a corresponding prompt message will be generated based on the specific judgment result of triggering termination, thereby helping the user to quickly identify the problem and adjust the registration environment.

[0143] In an exemplary embodiment, if registration is terminated because the noise energy ratio of the current speech segment exceeds a first threshold, the prompt message may be displayed as "The ambient noise is too high. Please try again in a quiet environment." If registration is terminated because the voiceprint similarity between the current speech segment and the previous speech segment is lower than a second threshold, the prompt message may be displayed as "Non-target user voice detected. Please ensure that only your own voice is being used." The above prompt messages can be output through the device's display screen, voice broadcast module, or mobile application interface, and may be in the form of at least one of text, icons, or synthesized speech.

[0144] By using the above methods, users can clearly distinguish whether the registration failure is caused by noise interference or other people's voices, thus improving the registration conditions in a targeted manner, avoiding blindly repeating operations, and significantly improving the registration success rate and interactive experience.

[0145] The voiceprint registration method of this disclosure is explained in detail below with a specific example. In this example, the voiceprint registration method is executed by a voiceprint registration system, which includes an environment detection and judgment module. When the user clicks the registration button, the voiceprint registration system is triggered to start; the system receives the continuous voice input by the user and divides it into multiple 3-second voice segments. The environment detection and judgment module is activated during the registration process, receiving one 3-second voice segment at a time (where the first 3-second voice segment is ignored), and performing noise estimation and voiceprint information extraction operations sequentially on subsequent voice segments every 3 seconds.

[0146] For the noise estimation and judgment part: starting from the second 3-second speech segment, the system determines whether the current environment is suitable for registration based on the noise estimation results. The noise estimation is implemented through a convolutional neural network with noise reduction function, which is a pre-trained model. During the training process, a large number of clean speech samples are mixed with noisy samples to simulate and generate noisy speech; using clean speech as the training target, the frequency domain features obtained by the simulated noisy speech after short-time Fourier transform (STFT) are used as the input of the convolutional neural network for end-to-end training, and finally the required noise reduction model is obtained.

[0147] like Figure 6As shown, in the noise estimation stage, the raw energy of the input 3-second speech segment is first calculated and denoted as s1. Then, this speech segment is input into the aforementioned noise reduction neural network, and the output is reconstructed into time-domain speech using inverse short-time Fourier transform (iSTFT), and its energy is calculated and denoted as s2. Further, the difference ratio ratio = (s1-s2) / s1 is calculated, and this ratio is used as the basis for judging the noise level. If the ratio is greater than the set noise threshold, the current environmental noise is determined to be too high and unsuitable for registration, and the system prompts a registration failure; if the ratio is less than or equal to the threshold, the subsequent process continues.

[0148] For the voiceprint information extraction and judgment part: such as Figure 7 As shown, the voiceprint information extraction process of the voiceprint feature extraction model is as follows: First, input a 3-second time-domain speech, extract the FBank (filter bank) features after STFT transformation, and then send the features into the voiceprint feature extraction model. The output is the voiceprint feature vector of the corresponding speech segment.

[0149] It should be noted that in the second 3-second voice segment, only voiceprint feature extraction is performed, and user consistency judgment is not performed at this time. From the third 3-second voice segment onwards, the system performs both noise estimation judgment and further processing of the extracted voiceprint features. Specifically, assuming the voiceprint feature of the previous 3-second voice segment is e1 and the voiceprint feature of the current 3-second voice segment is e2, the cosine similarity between the two is calculated as cossim. If cossim is not lower than the set similarity threshold, it is determined that the current voice segment and the previous voice segment come from the same user, and the process continues; if cossim is lower than the threshold, it is determined that there is non-target user voice, and the system prompts registration failure.

[0150] The noise estimation and voiceprint similarity judgment are related by an "AND" logic: if either condition is not met, registration terminates and a failure message is displayed. The noise threshold and voiceprint similarity threshold used in the flowchart can be adjusted according to the actual application scenario; the duration of each speech segment can also be modified as needed. This judgment process continues until the accumulated effective speech duration reaches the minimum speech duration required for registration, thus completing the voiceprint registration.

[0151] This embodiment combines a noise reduction model and a voiceprint feature extraction model to dynamically assess the compliance of the voice environment during registration. On the one hand, it quantifies the environmental noise level based on the energy difference ratio between the input and output of the noise reduction model, avoiding the problem of poor adaptability in different acoustic scenarios by relying on fixed preset thresholds. On the other hand, it effectively identifies the presence of non-target user voices through cross-speech segment voiceprint similarity comparison. The synergistic effect of these two aspects enables the system to accurately determine whether the current voiceprint registration conditions of "single user, low noise" are met in a real and dynamic usage environment, thereby significantly improving the purity of the registered voice and the reliability of the voiceprint model.

[0152] It should be noted that the above figures are merely illustrative representations of the processes included in methods according to some embodiments of this disclosure, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0153] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein. For details not disclosed in the apparatus embodiments of this disclosure, please refer to the embodiments of the method disclosed herein.

[0154] Figure 8 This is a block diagram illustrating a voiceprint registration device according to some embodiments of the present disclosure. (Refer to...) Figure 8 The device includes: a voice receiving module 801, a voice processing module 802, and a voiceprint registration module 803.

[0155] The voice receiving module 801 is used to receive voice information in response to receiving a user's voiceprint registration command; The voice processing module 802 is used to divide voice information into multiple voice segments; The voiceprint registration module 803 is used to register voiceprints based on voice information in response to the fact that the environmental noise levels of multiple voice segments do not exceed a first threshold and any consecutive voice segments correspond to the same user.

[0156] In some embodiments of this disclosure, the voiceprint registration module is used to calculate the total duration of multiple voice segments in response to the fact that the environmental noise levels of multiple voice segments do not exceed a first threshold and any consecutive voice segments correspond to the same user; and to perform voiceprint registration based on multiple voice segments in response to the total duration reaching a preset duration threshold.

[0157] In some embodiments of this disclosure, the voiceprint registration device further includes: a noise judgment module, configured to convert the current speech segment into frequency domain data; perform noise reduction processing on the frequency domain data to obtain noise-reduced frequency domain data; convert the noise-reduced frequency domain data back to the time domain to obtain noise-reduced speech; calculate the original energy of the current speech segment and the energy of the noise-reduced speech respectively; divide the difference between the original energy and the energy of the noise-reduced speech by the original energy to obtain a noise energy ratio; and determine that the environmental noise level of the current speech segment exceeds the first threshold when the noise energy ratio is greater than a first threshold.

[0158] In some embodiments of this disclosure, the voiceprint registration device further includes: a user judgment module, configured to extract voiceprint features of the current speech segment using a voiceprint feature extraction model, and extract voiceprint features of the previous speech segment using a voiceprint feature extraction model; calculate the similarity between the voiceprint features of the current speech segment and the voiceprint features of the previous speech segment; and determine that the current speech segment and the previous speech segment originate from the same user when the similarity is not lower than a second threshold.

[0159] In some embodiments of this disclosure, the user judgment module extracts the voiceprint features of the current speech segment using a voiceprint feature extraction model, including: performing a short-time Fourier transform on the current speech segment to obtain the frequency domain data of the current speech segment; calculating the power spectrum based on the frequency domain data, and extracting filter bank features based on the power spectrum; and inputting the filter bank features into the voiceprint feature extraction model to obtain the voiceprint features of the current speech segment.

[0160] In some embodiments of this disclosure, the speech processing module 802 is used to divide continuous speech in speech information into multiple speech segments according to a fixed duration.

[0161] In some embodiments of this disclosure, the voiceprint registration device further includes: a prompt output module, configured to output a prompt message to the user in response to the ambient noise level of any voice segment exceeding the first threshold, or any consecutive voice segments corresponding to different users, wherein the prompt message indicates that the registration failure is due to excessive ambient noise or the detection of non-target user voice.

[0162] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the voiceprint registration method, and will not be elaborated upon here.

[0163] This disclosure also provides a vehicle, which can be a hybrid vehicle, a non-hybrid vehicle, an electric vehicle, a fuel cell vehicle, or other types of vehicles. The vehicle can be a driver-assisted vehicle, a semi-driver-assisted vehicle, or a driver-free vehicle.

[0164] Vehicles can include various subsystems, such as infotainment systems, perception systems, decision control systems, drive systems, and computing platforms. A vehicle can also include more or fewer subsystems, and each subsystem can include multiple components. Furthermore, each subsystem and each component of the vehicle can be interconnected via wired or wireless means.

[0165] In some embodiments, an infotainment system may include a communication system, an entertainment system, and a navigation system, etc.

[0166] The perception system may include several types of sensors used to sense information about the environment surrounding the vehicle. For example, the perception system may include a global positioning system (which may be GPS, BeiDou, or other positioning systems), an inertial measurement unit (IMU), lidar, millimeter-wave radar, ultrasonic radar, and camera devices.

[0167] The decision control system may include a computing system, a vehicle controller, a steering system, a throttle, and a braking system.

[0168] A drive system may include components that provide powered motion to a vehicle. In one embodiment, a drive system may include an engine, an energy source, a transmission system, and wheels. The engine may be one or a combination of internal combustion engines, electric motors, and compressed air engines. The engine is capable of converting energy provided by the energy source into mechanical energy.

[0169] Some or all of the vehicle's functions are controlled by a computing platform. The computing platform may include at least one processor and memory, the processor being able to execute instructions stored in the memory.

[0170] The processor can be any conventional processor, such as a commercially available CPU. The processor can also include graphics processing units (GPUs), field-programmable gate arrays (FPGAs), systems-on-chips (SoCs), application-specific integrated circuits (ASICs), or combinations thereof.

[0171] Memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0172] In addition to instructions, memory can also store data, such as road maps, route information, and vehicle position, direction, and speed. The data stored in memory can be used by the computing platform.

[0173] In this embodiment of the disclosure, the processor can execute instructions to complete all or part of the steps of the above-described voiceprint registration method.

[0174] Figure 9 This is a block diagram illustrating an electronic device 900 according to some embodiments of the present disclosure. For example, the electronic device 900 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0175] Reference Figure 9 The electronic device 900 may include one or more of the following components: a processing component 902, a memory 904, a power component 906, a multimedia component 908, an audio component 910, an input / output (I / O) interface 912, a sensor component 914, and a communication component 916.

[0176] Processing component 902 typically controls the overall operation of electronic device 900, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 902 may include one or more processors 920 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 902 may include one or more modules to facilitate interaction between processing component 902 and other components. For example, processing component 902 may include a multimedia module to facilitate interaction between multimedia component 908 and processing component 902.

[0177] Memory 904 is configured to store various types of data to support the operation of device 900. Examples of this data include instructions for any application or method operating on electronic device 900, contact data, phonebook data, messages, pictures, videos, etc. Memory 904 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0178] Power supply component 906 provides power to various components of electronic device 900. Power supply component 906 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 900.

[0179] Multimedia component 908 includes a screen that provides an output interface between the electronic device 900 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 908 includes a front-facing camera and / or a rear-facing camera. When the device 900 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0180] Audio component 910 is configured to output and / or input audio signals. For example, audio component 910 includes a microphone (MIC) configured to receive external audio signals when electronic device 900 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 904 or transmitted via communication component 916. In some embodiments, audio component 910 also includes a speaker for outputting audio signals.

[0181] I / O interface 912 provides an interface between processing component 902 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0182] Sensor assembly 914 includes one or more sensors for providing state assessments of various aspects of electronic device 900. For example, sensor assembly 914 can detect the on / off state of device 900, the relative positioning of components such as the display and keypad of electronic device 900, changes in position of electronic device 900 or a component of electronic device 900, the presence or absence of user contact with electronic device 900, orientation or acceleration / deceleration of electronic device 900, and temperature changes of electronic device 900. Sensor assembly 914 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 914 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 914 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0183] Communication component 916 is configured to facilitate wired or wireless communication between electronic device 900 and other devices. Electronic device 900 can access wireless networks based on communication standards, such as WiFi, 3G, 4G, 5G, other communication standards, or combinations thereof. In some embodiments of this disclosure, communication component 916 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In some embodiments of this disclosure, communication component 916 further includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0184] In some embodiments of this disclosure, the electronic device 900 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described voiceprint registration method.

[0185] In some embodiments of this disclosure, a computer-readable storage medium including instructions is also provided, such as a memory 904 including instructions, which can be executed by a processor 920 of an electronic device 900 to perform the above-described method. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0186] This disclosure also provides a computer-readable storage medium storing computer instructions thereon, which, when executed by a processor, implement the voiceprint registration method described in the above method embodiments.

[0187] In this disclosure, the computer-readable storage medium is one capable of sending, propagating, or transmitting computer instructions for use by or in connection with an instruction execution system, apparatus, or device. As an example, the computer-readable storage medium is a non-volatile storage medium.

[0188] In some embodiments, more specific examples of computer-readable storage media in this disclosure may include, but are not limited to: electrical connections having 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 fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, USB flash drives, portable hard drives, or any suitable combination of the foregoing.

[0189] In some examples, computational instructions contained on a computer-readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0190] This disclosure also provides a computer program product storing instructions that, when executed by a computer, cause the computer to implement the voiceprint registration method described in the above-described method embodiments. These instructions may be program code. In specific implementations, the program code may be written using any combination of one or more programming languages. The program code may execute entirely on a user's computing device, partially on a user's device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0191] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0192] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A voiceprint registration method, characterized in that, include: In response to receiving a user's voiceprint registration instruction, receive voice information; The voice information is divided into multiple voice segments; In response to the fact that the environmental noise levels of the multiple voice segments do not exceed a first threshold, and that any consecutive voice segments correspond to the same user, voiceprint registration is performed based on the voice information.

2. The method according to claim 1, characterized in that, The response that the environmental noise levels of the multiple speech segments do not exceed a first threshold, and that any consecutive speech segments correspond to the same user, includes voiceprint registration based on the speech information, comprising: In response to the fact that the environmental noise levels of the multiple voice segments do not exceed the first threshold and any consecutive voice segments correspond to the same user, the total duration of the multiple voice segments is calculated. In response to the total duration reaching a preset duration threshold, voiceprint registration is performed based on the multiple speech segments.

3. The method according to claim 1, characterized in that, The method further includes: Convert the current speech segment into frequency domain data; The frequency domain data is denoised to obtain denoised frequency domain data. The denoised frequency domain data is converted back to the time domain to obtain the denoised speech; Calculate the original energy of the current speech segment and the energy of the denoised speech segment respectively; The difference between the original energy and the energy of the noise-reduced speech is divided by the original energy to obtain the noise energy ratio. When the noise energy ratio is greater than the first threshold, it is determined that the environmental noise level of the current speech segment exceeds the first threshold.

4. The method according to claim 1, characterized in that, The method further includes: The voiceprint feature extraction model is used to extract the voiceprint features of the current speech segment, and the voiceprint feature extraction model is used to extract the voiceprint features of the previous speech segment. Calculate the similarity between the voiceprint features of the current speech segment and the voiceprint features of the previous speech segment; When the similarity is not lower than the second threshold, it is determined that the current voice segment and the previous voice segment originate from the same user.

5. The method according to claim 4, characterized in that, The process of extracting the voiceprint features of the current speech segment using a voiceprint feature extraction model includes: Perform a short-time Fourier transform on the current speech segment to obtain the frequency domain data of the current speech segment; The power spectrum is calculated based on the frequency domain data, and filter bank features are extracted based on the power spectrum. The filter bank features are input into the voiceprint feature extraction model to obtain the voiceprint features of the current speech segment.

6. The method according to any one of claims 1-5, characterized in that, The step of dividing the voice information into multiple voice segments includes: The continuous speech in the speech information is divided into multiple speech segments according to a fixed duration.

7. The method according to claim 1, characterized in that, The method further includes: In response to the ambient noise level of any voice segment exceeding the first threshold, or any consecutive voice segments corresponding to different users, a prompt message is output to the user, indicating that the registration failure is due to excessive ambient noise or the detection of non-target user voice.

8. A voiceprint registration device, characterized in that, include: The voice receiving module is used to receive voice information in response to receiving a user's voiceprint registration command; The voice processing module is used to divide the voice information into multiple voice segments; The voiceprint registration module is used to register voiceprints based on the voice information when the environmental noise levels of the multiple voice segments do not exceed a first threshold and any consecutive voice segments correspond to the same user.

9. The apparatus according to claim 8, characterized in that, The device further includes: A noise detection module is used to convert the current speech segment into frequency domain data; perform noise reduction processing on the frequency domain data to obtain noise-reduced frequency domain data; convert the noise-reduced frequency domain data back to the time domain to obtain noise-reduced speech; calculate the original energy of the current speech segment and the energy of the noise-reduced speech respectively; divide the difference between the original energy and the energy of the noise-reduced speech by the original energy to obtain a noise energy ratio; when the noise energy ratio is greater than a first threshold, determine that the environmental noise level of the current speech segment exceeds the first threshold.

10. The apparatus according to claim 8, characterized in that, The device further includes: The user determination module is used to extract the voiceprint features of the current speech segment using a voiceprint feature extraction model, and to extract the voiceprint features of the previous speech segment using the same voiceprint feature extraction model; calculate the similarity between the voiceprint features of the current speech segment and the voiceprint features of the previous speech segment; and determine that the current speech segment and the previous speech segment originate from the same user when the similarity is not lower than a second threshold.

11. The apparatus according to claim 10, characterized in that, The user judgment module extracts the voiceprint features of the current speech segment using a voiceprint feature extraction model, including: performing a short-time Fourier transform on the current speech segment to obtain the frequency domain data of the current speech segment; calculating the power spectrum based on the frequency domain data, and extracting filter bank features based on the power spectrum; and inputting the filter bank features into the voiceprint feature extraction model to obtain the voiceprint features of the current speech segment.

12. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to implement the steps of the voiceprint registration method according to any one of claims 1-7.

13. A computer-readable storage medium, wherein instructions in the storage medium, when executed by a processor of a mobile terminal, enable the mobile terminal to perform the steps of the voiceprint registration method according to any one of claims 1-7.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the voiceprint registration method as described in any one of claims 1-7.

Images