A vocal pitch detection method, device, equipment and medium
By combining deep learning and a dedicated smoothing algorithm with closed-loop feedback optimization, the problem of misjudging artistic singing techniques in existing technologies has been solved, achieving high-precision and robust pitch detection and improving the accuracy and practicality of the detection results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- MINJIANG NORMAL COLLEGE
- Filing Date
- 2026-03-10
- Publication Date
- 2026-05-29
AI Technical Summary
Existing pitch detection methods often misjudge or mask vocal techniques when dealing with artistic singing skills due to the use of a uniform smoothing strategy, affecting the accuracy and reliability of the detection results.
We employ a deep learning-based automatic marking mechanism for singing techniques, combined with dedicated smoothing algorithms for different techniques (such as vibrato, glissando, and trills), and dynamically adjust the recognition threshold and feature weights by constructing a closed-loop feedback optimization path from pitch smoothing processing results to the technique marking model.
It significantly improves the accuracy and robustness of pitch trajectory analysis, effectively distinguishes between artistic singing fluctuations and pitch errors, provides high-precision pitch deviation extraction, and enhances the system's generalization ability to different users and complex singing scenarios.
Smart Images
Figure CN122116856A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of music information retrieval and computer-aided music, specifically a method, device, equipment and medium for vocal pitch detection. Background Technology
[0002] Vocal pitch detection is a core technology in music information retrieval and computer-aided music education. Its goal is to automate and objectively analyze and evaluate pitch information in vocal audio. Pitch accuracy is a fundamental indicator of vocal performance quality. Traditional and mainstream pitch detection methods typically follow a standard procedure: first, preprocessing the original audio including endpoint detection, noise reduction, and framing; then, extracting fundamental frequency features to generate a pitch sequence; and finally, comparing this sequence with a standard musical score to calculate pitch deviation. Robust pitch trajectory estimation is a crucial step in ensuring the reliability of the final evaluation results.
[0003] In actual artistic performance, singers often employ various singing techniques to enhance expressiveness, such as vibrato, glissando, and trills. These techniques manifest acoustically as specific patterns of pitch variation: vibrato is a regular, periodic fluctuation of pitch around a steady-state value; glissando is a smooth, continuous transition of pitch between two different notes; and trills involve switching between rapidly passing notes. These controlled, artistic pitch variations, along with unexpected, irregular pitch deviations caused by singing errors or breath instability, often intertwine at the signal level. This constitutes one of the core challenges facing pitch detection technology: how to accurately distinguish between artistic, music-theoretical pitch fluctuations and genuine pitch errors.
[0004] Most existing pitch detection methods have significant limitations in addressing this problem. A common approach is to ignore the differences in singing technique and uniformly apply a general smoothing filter algorithm (such as median filtering, mean filtering, or low-pass filtering) to all pitch sequences to remove noise and outliers. However, this "one-size-fits-all" approach has inherent flaws: the parameter settings of general filters are usually fixed, designed to suppress all rapid changes. This means that while filtering out unexpected deviations, it may also inappropriately suppress or even erase rapid, technical pitch changes that conform to artistic principles (such as rapid glissando or trills), or disrupt the natural fluctuation pattern of vibrato, resulting in distorted pitch trajectories. Ultimately, pitch deviation assessments based on distorted trajectories will lose accuracy, potentially misjudging correct technical performances as off-key, or masking genuine pitch problems, making the test results unable to truly reflect the singer's pitch level and limiting the application value of this technology in professional vocal teaching and assessment. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a vocal pitch detection method, apparatus, device, and medium to resolve the issue in the prior art where the use of a uniform pitch smoothing strategy leads to the misjudgment or masking of artistic singing techniques, thereby affecting the accuracy and reliability of pitch detection results.
[0006] A method for detecting vocal pitch includes the following steps:
[0007] S1: Acquire and preprocess the raw audio signal to obtain the audio segment to be analyzed;
[0008] S2: Extract the basic acoustic features of the audio segment to be analyzed;
[0009] S3: Based on the aforementioned basic acoustic features, perform pitch trajectory analysis to obtain the target pitch sequence;
[0010] S4: Compare the target pitch sequence with the standard pitch reference and calculate the pitch deviation.
[0011] S5: Output the detection result including the pitch deviation;
[0012] Step S1 includes:
[0013] S11: Perform endpoint detection on the original audio signal to determine valid human voice segments;
[0014] S12: Perform noise reduction and normalization processing on the effective human voice segments;
[0015] S13: Perform frame segmentation and windowing operations on the processed audio to generate the audio segment to be analyzed;
[0016] S14: Perform the first feature marking to mark a preset specific singing technique segment in the audio segment to be analyzed. The specific singing technique segment includes at least one of vibrato segment, glissando segment, or trill segment.
[0017] Step S3 includes:
[0018] S31: Based on the aforementioned fundamental acoustic features, an initial pitch sequence is generated using a fundamental frequency estimation algorithm;
[0019] S32: Perform a first smoothing process on the initial pitch sequence to eliminate abnormal jump points caused by short-term noise;
[0020] S33: Based on the first feature marker, apply a second smoothing process to the pitch sequence segment corresponding to the specific singing technique segment, wherein the smoothing window and / or smoothing algorithm used in the second smoothing process are different from the first smoothing process;
[0021] S34: The pitch sequence after the first smoothing process and / or the second smoothing process is taken as the target pitch sequence;
[0022] In step S14, the labeling result of the specific singing technique segment is directly input into step S33 as a control parameter to activate and configure the second smoothing process; and the output result of the second smoothing process in step S33 is fed back to the labeling model used in step S14 to update the recognition threshold or feature weight of the labeling model for the same type of singing technique segment.
[0023] Preferably, the basic acoustic features extracted in step S2 include Mel frequency cepstral coefficients, spectral centroid, short-time energy, and zero-crossing rate.
[0024] Preferably, in step S33, the second smoothing process is a dynamic smoothing threshold process, specifically including:
[0025] When the first feature is marked as a vibrato segment, a bandpass filter adapted to the vibrato frequency is used for smoothing.
[0026] When the first feature is marked as a glissando segment, constraint smoothing based on the first derivative is used;
[0027] When the first feature is marked as a pitch transition, a step-like smoothing based on note segmentation is used.
[0028] Preferably, step S4 includes:
[0029] S41: Based on the song information, load the corresponding sheet music information as the standard pitch reference, wherein the sheet music information includes the standard pitch value, duration and note boundary;
[0030] S42: Using a dynamic time warping algorithm, the target pitch sequence is time-aligned with the standard pitch reference;
[0031] S43: Calculate the instantaneous pitch deviation at each time point after alignment, and statistically analyze at least one of the mean, variance, and maximum deviation of the overall pitch deviation to obtain the pitch deviation degree.
[0032] Preferably, in step S5, the detection result is output in the form of a visual pitch curve comparison chart, wherein the target pitch sequence and the standard pitch reference are plotted on the same coordinate system, and the segments that exceed the preset deviation threshold are highlighted.
[0033] A vocal pitch detection device, comprising:
[0034] The audio preprocessing module is configured to perform step S1 as described above.
[0035] The feature extraction module is configured to perform step S2 of the method described above;
[0036] The pitch analysis module is configured to perform step S3 as described above.
[0037] The pitch evaluation module is configured to perform step S4 as described above.
[0038] The result output module is configured to perform step S5 of the method described above.
[0039] Preferably, the audio preprocessing module includes:
[0040] Endpoint detection unit, used to perform step S11;
[0041] The signal enhancement unit is used to perform step S12;
[0042] The framing and windowing unit is used to execute step S13;
[0043] The skill tagging unit is used to perform step S14 and contains a tagging model that can be updated based on feedback.
[0044] Preferably, the pitch analysis module includes:
[0045] The initial estimation unit is used to perform step S31;
[0046] A general-purpose smoothing unit is used to perform step S32;
[0047] An adaptive smoothing unit, used to perform step S33, receives a control signal from the trick mark unit;
[0048] A sequence generation unit is used to perform step S34.
[0049] An electronic device, comprising:
[0050] One or more processors;
[0051] Storage device for storing one or more programs;
[0052] When the one or more programs are executed by the one or more processors, the one or more processors implement the vocal pitch detection method described above.
[0053] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the vocal pitch detection method described above.
[0054] Compared with the prior art, the present invention has the following beneficial effects:
[0055] By introducing a deep learning-based automatic labeling mechanism for singing techniques and combining it with dedicated smoothing algorithms for different techniques (such as vibrato, glissando, and trills), the accuracy and robustness of pitch trajectory analysis are significantly improved. This enables the system to effectively distinguish between artistic singing fluctuations and pitch errors, avoiding the problem of misjudging or erasing singing techniques due to the "one-size-fits-all" smoothing of traditional methods. Thus, it achieves accurate extraction of true pitch deviations while preserving artistic expressiveness.
[0056] By constructing a closed-loop feedback optimization path from the pitch smoothing processing results to the technique labeling model, the system possesses continuous self-learning and adaptive evolution capabilities. The labeling model can dynamically adjust its internal recognition threshold and feature weights based on the feedback from the smoothing effect, thereby continuously optimizing the recognition accuracy for different singing styles and technique variations in practical use. This mechanism effectively reduces the system's dependence on preset parameters and enhances its generalization ability to different users and complex singing scenarios.
[0057] By integrating the aforementioned innovative algorithms into dedicated hardware and storage media, a complete transformation from algorithmic theory to practical products has been achieved. The dedicated equipment ensures real-time processing through an optimized heterogeneous computing architecture, while the integrated interactive design greatly enhances the user experience. This allows highly professional pitch detection technology to move from laboratory environments to everyday teaching and practice scenarios, providing vocal learners with efficient, intuitive, and personalized tools for accurate feedback, truly realizing the empowerment of arts education through technology. Attached Figure Description
[0058] Figure 1 This is a flowchart illustrating the vocal pitch detection method of the present invention. Detailed Implementation
[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.
[0060] like Figure 1 As shown:
[0061] Example: Specific Implementation of a Vocal Pitch Detection Method
[0062] This embodiment provides a specific implementation scheme for a vocal pitch detection method, aiming to explain in detail the implementation details and logic of each step.
[0063] S1: Acquire and preprocess the raw audio signal to obtain the audio segment to be analyzed.
[0064] The specific implementation is as follows:
[0065] Use microphone arrays or high-precision recording equipment to capture the singer's raw audio signal. .
[0066] The acquired raw audio signal is preprocessed to generate the audio segment to be analyzed, which includes the following sub-steps:
[0067] S11: Endpoint detection. For the raw audio signal Dual-threshold endpoint detection based on short-time energy and zero-crossing rate is performed to determine the starting point of the valid human voice segment. and end point This allows for the extraction of effective vocal segments. ,in .
[0068] S12: Noise Reduction and Normalization. For effective human voice segments. Spectral subtraction is applied for noise reduction to eliminate steady-state environmental noise. Subsequently, the denoised signal undergoes amplitude normalization to unify its peak amplitude to a preset value. Receive signal .
[0069] S13: Framing and Windowing. This involves framing the normalized audio signal... Frames are processed sequentially by time, with a frame length of [length missing]. (Typically 20-40ms), frame shift is (Typically 1 / 2 of the frame length). For each frame signal Using the Hamming window function Windowing is applied to obtain the windowed frame signal. ,in The frame number is the number of the frames. The set of all frames constitutes the audio segment to be analyzed. .
[0070] S14: Perform the first feature labeling. Based on the temporal variation characteristics of the Mel frequency cepstral coefficients and spectral centroid, a pre-trained convolutional neural network model is used to analyze the audio segment. The analysis identifies and marks segments that contain specific singing techniques. Specifically, segments containing specific singing techniques include vibrato segments. glissando or transitional phrase At least one of them. The labeling result is output in the form of a time-series label vector, denoted as . , used to indicate the The skill category to which frame audio belongs.
[0071] S2: Extract the basic acoustic features of the audio segment to be analyzed.
[0072] For each frame of windowed signal The following basic acoustic features are extracted in parallel to form the feature vector of this frame. :
[0073] Mel frequency cepstral coefficients: calculation The MFCC coefficients, taking the first 13 dimensions (including the 0th-order coefficients), are denoted as... .
[0074] Spectral centroid: Calculation Spectral centroid As .
[0075] Short-time energy: calculation short-term energy As .
[0076] Zero crossing rate: Calculation Zero crossing rate As .
[0077] By concatenating the above features, we obtain the complete feature vector for each frame. The set of feature vectors from all frames constitutes the basic acoustic feature sequence for subsequent analysis.
[0078] S3: Based on the basic acoustic features, perform pitch trajectory analysis to obtain the target pitch sequence.
[0079] The specific implementation is as follows:
[0080] S31: Generate the initial pitch sequence. This is done for the feature vector of each frame. The initial fundamental frequency of the frame is calculated by using the YIN fundamental frequency estimation algorithm, focusing on its spectral characteristics. To form the initial pitch sequence .
[0081] S32: First smoothing process. For the initial pitch sequence... A median filter is applied for the first smoothing process. A fixed-width sliding window (e.g., corresponding to a duration of 100ms) is set, and the fundamental frequency values are sorted and the median is taken within the window to eliminate short-term anomalous jumps caused by pops or transient noise, thus obtaining an intermediate sequence. .
[0082] S33: Second smoothing process based on tags. Read the first feature tags from step S14. Based on the labels, the sequence For sections belonging to specific singing techniques, a second smoothing process adapted to the technique is applied:
[0083] like Marked as a trill segment Adaptive bandpass filtering based on vibrato frequency is used for smoothing. First, the fluctuation frequency of the pitch range is detected. Then design a center frequency of Bandwidth is The bandpass filter removes abnormal components outside the frequency range of the oscillation, preserving the natural tremolo oscillation pattern.
[0084] like Marked as glissando Constrained smoothing based on the first derivative is employed. The first-order difference (derivative) of the pitch range is calculated, and a maximum rate of change threshold is set for the difference value. For any changes exceeding the threshold, adjust them to the maximum change value within the allowable range of the threshold to ensure the continuity and smoothness of the glissando process and avoid abrupt, step-like jumps.
[0085] like Marked as a transition segment The method employs a step-like smoothing approach based on note segmentation. First, a note start point detection algorithm is used to segment the target note intervals within the transition segment. Then, the pitch values within each interval are smoothed to the stable mean or median of the pitch within that interval, forming a step-like pitch trajectory with clear and smooth transitions between different notes.
[0086] S34: Generate the target pitch sequence. The pitch values processed in S32 and S33 are integrated to form the final target pitch sequence. .
[0087] Feedback Update Mechanism: The smoothness and continuity indices of the pitch trajectory after the second smoothing process in step S33 will be used as feedback signals to be transmitted back to the labeling model used in step S14. The model uses these feedback signals to fine-tune its internal parameters through the backpropagation algorithm, thereby updating the recognition thresholds or feature weights for techniques such as vibrato, glissando, and trills, achieving an adaptive improvement in labeling accuracy.
[0088] S4: Compare the target pitch sequence with the standard pitch reference and calculate the pitch deviation.
[0089] The specific implementation is as follows:
[0090] S41: Load Standard Pitch Reference. Based on the input song title, load the corresponding digital sheet music information from the standard sheet music database as the standard pitch reference. This reference information includes the standard pitch value (in Hz), start time, duration, and note boundaries for each note.
[0091] S42: Time Alignment. Utilizing a dynamic time warping algorithm, the target pitch sequence is aligned... (Timeline is frame number) ) and standard pitch reference Non-linear time alignment is performed to eliminate the time difference between the singing rhythm and the standard rhythm, and to establish the correspondence between frames and reference time points.
[0092] S43: Calculate pitch deviation. Based on time alignment, for each aligned time point... Calculate instantaneous pitch deviation (Usually converted to cents). Then, the following indicators for the entire piece are statistically analyzed to comprehensively evaluate the pitch deviation:
[0093] Mean absolute deviation:
[0094] Deviation Standard Deviation:
[0095] Maximum deviation:
[0096] pitch deviation It can be represented by a weighted combination of one or more of the above indicators, for example .
[0097] S5: Outputs the detection results including pitch deviation.
[0098] Generate and output a visualized detection report. The core part of the report is a pitch curve comparison chart, which plots the target pitch sequence on the same time coordinate system. The trajectory curve and standard pitch reference A stepped reference line. Different colors or line styles are used to distinguish between the two. Simultaneously, instantaneous deviations are plotted on the graph. Exceeding the preset deviation threshold Sections (e.g., 50 cents) are highlighted with shading. The report also lists in text form the various pitch deviation indices calculated in S43. The specific numerical values are provided, along with an overall pitch rating assessment (e.g., excellent, good, requires practice, etc.).
[0099] This embodiment achieves high-precision and robust pitch detection of vocal audio containing complex singing techniques by deeply coupling specific singing skills identification (first feature labeling) with differentiated pitch smoothing strategies (second smoothing processing) and introducing a feedback optimization mechanism, effectively improving the accuracy and practicality of the detection results.
[0100] Example 2: This example provides a vocal pitch detection device, including:
[0101] The audio preprocessing module is configured to perform step S1 as described above.
[0102] The feature extraction module is configured to perform step S2 of the method described above;
[0103] The pitch analysis module is configured to perform step S3 as described above.
[0104] The pitch evaluation module is configured to perform step S4 as described above.
[0105] The result output module is configured to execute step S5 of the method described above.
[0106] Specifically, the electronic device is a dedicated vocal teaching and assessment all-in-one machine, which is a portable device with a high-quality microphone, display screen and physical buttons.
[0107] The device hardware components include:
[0108] Central Processing Unit: Employs a high-performance octa-core processor with a clock speed of 2.8GHz, and is equipped with a dedicated neural network processing unit to accelerate real-time inference of the skill tagging model.
[0109] Audio processing subsystem:
[0110] Two professional-grade condenser microphones with a sensitivity of -35dB and a frequency response of 20Hz-20kHz.
[0111] 24-bit high-precision analog-to-digital converter, supporting a 192kHz sampling rate;
[0112] An independent noise reduction chip eliminates ambient noise in real time;
[0113] Storage device: Includes a 256GB solid-state drive (SSD) as main storage and 8GB of high-speed RAM. The SSD is divided into multiple logical partitions:
[0114] System program area: Stores the operating system kernel and the main program for vocal pitch detection;
[0115] Model parameter area: Stores parameters of pre-trained deep learning models, including skill tagging models, pitch detection models, etc.
[0116] Standard Song Library: Stores MIDI data and sheet music information for over 1000 standard songs;
[0117] User data area: Stores the user's testing history, personalized settings, and practice records;
[0118] Display and Interaction Unit:
[0119] 10.1-inch fully laminated IPS touchscreen with a resolution of 1920×1200;
[0120] Six dedicated function buttons: Record / Stop, Play, Evaluate, Practice, Settings, Save;
[0121] Rotary encoder, used to adjust volume and sensitivity;
[0122] Communication interface:
[0123] The USB-C interface supports audio input / output and data transfer.
[0124] Bluetooth 5.2 module, which can connect to wireless headphones or speakers;
[0125] Wi-Fi 6 module, supports online music library updates and cloud synchronization;
[0126] Specifically, the audio preprocessing module includes:
[0127] Endpoint detection unit, used to perform step S11;
[0128] The signal enhancement unit is used to perform step S12;
[0129] The framing and windowing unit is used to execute step S13;
[0130] The skill tagging unit is used to perform step S14 and contains a tagging model that can be updated based on feedback.
[0131] Specifically, the pitch analysis module includes:
[0132] The initial estimation unit is used to perform step S31;
[0133] A general-purpose smoothing unit is used to perform step S32;
[0134] An adaptive smoothing unit, used to perform step S33, receives a control signal from the trick mark unit;
[0135] A sequence generation unit is used to perform step S34.
[0136] An electronic device, comprising:
[0137] One or more processors;
[0138] Storage device for storing one or more programs;
[0139] When one or more programs are executed by one or more processors, the one or more processors implement the vocal pitch detection method described above.
[0140] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the vocal pitch detection method described above.
[0141] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0142] The accompanying drawings of the embodiments disclosed in this invention only involve structures relevant to the embodiments disclosed in this invention. Other structures can be referred to with common designs. Unless otherwise specified, the same embodiment and different embodiments of this invention can be combined with each other.
[0143] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for detecting vocal pitch, characterized in that, Includes the following steps: S1: Acquire and preprocess the raw audio signal to obtain the audio segment to be analyzed; S2: Extract the basic acoustic features of the audio segment to be analyzed; S3: Based on the aforementioned basic acoustic features, perform pitch trajectory analysis to obtain the target pitch sequence; S4: Compare the target pitch sequence with the standard pitch reference and calculate the pitch deviation. S5: Output the detection result including the pitch deviation; Step S1 includes: S11: Perform endpoint detection on the original audio signal to determine valid human voice segments; S12: Perform noise reduction and normalization processing on the effective human voice segments; S13: Perform frame segmentation and windowing operations on the processed audio to generate the audio segment to be analyzed; S14: Perform the first feature marking to mark a preset specific singing technique segment in the audio segment to be analyzed. The specific singing technique segment includes at least one of vibrato segment, glissando segment, or trill segment. Step S3 includes: S31: Based on the aforementioned fundamental acoustic features, an initial pitch sequence is generated using a fundamental frequency estimation algorithm; S32: Perform a first smoothing process on the initial pitch sequence to eliminate abnormal jump points caused by short-term noise; S33: Based on the first feature marker, apply a second smoothing process to the pitch sequence segment corresponding to the specific singing technique segment, wherein the smoothing window and / or smoothing algorithm used in the second smoothing process are different from the first smoothing process; S34: The pitch sequence after the first smoothing process and / or the second smoothing process is taken as the target pitch sequence; In step S14, the labeling result of the specific singing technique segment is directly input into step S33 as a control parameter to activate and configure the second smoothing process; and the output result of the second smoothing process in step S33 is fed back to the labeling model used in step S14 to update the recognition threshold or feature weight of the labeling model for the same type of singing technique segment.
2. The vocal pitch detection method according to claim 1, characterized in that, The basic acoustic features extracted in step S2 include Mel frequency cepstral coefficients, spectral centroid, short-time energy, and zero-crossing rate.
3. The vocal pitch detection method according to claim 1, characterized in that, In step S33, the second smoothing process is a dynamic smoothing threshold process, specifically including: When the first feature is marked as a vibrato segment, a bandpass filter adapted to the vibrato frequency is used for smoothing. When the first feature is marked as a glissando segment, constraint smoothing based on the first derivative is used; When the first feature is marked as a pitch transition, a step-like smoothing based on note segmentation is used.
4. The vocal pitch detection method according to claim 1, characterized in that, Step S4 includes: S41: Based on the song information, load the corresponding sheet music information as the standard pitch reference, wherein the sheet music information includes the standard pitch value, duration and note boundary; S42: Using a dynamic time warping algorithm, the target pitch sequence is time-aligned with the standard pitch reference; S43: Calculate the instantaneous pitch deviation at each time point after alignment, and statistically analyze at least one of the mean, variance, and maximum deviation of the overall pitch deviation to obtain the pitch deviation degree.
5. The vocal pitch detection method according to claim 1, characterized in that, In step S5, the detection results are output in the form of a visual pitch curve comparison chart, wherein the target pitch sequence and the standard pitch reference are plotted on the same coordinate system, and the segments that exceed the preset deviation threshold are highlighted.
6. A vocal pitch detection device, characterized in that, include: The audio preprocessing module is configured to perform step S1 of the method as described in any one of claims 1 to 5; The feature extraction module is configured to perform step S2 of the method as described in any one of claims 1 to 5; The pitch analysis module is configured to perform step S3 of the method as described in any one of claims 1 to 5; The pitch evaluation module is configured to perform step S4 of the method as described in any one of claims 1 to 5; The result output module is configured to perform step S5 of the method as described in any one of claims 1 to 5.
7. The apparatus according to claim 6, characterized in that, The audio preprocessing module includes: Endpoint detection unit, used to perform step S11; The signal enhancement unit is used to perform step S12; The framing and windowing unit is used to execute step S13; The skill tagging unit is used to perform step S14 and contains a tagging model that can be updated based on feedback.
8. The apparatus according to claim 6, characterized in that, The pitch analysis module includes: The initial estimation unit is used to perform step S31; A general-purpose smoothing unit is used to perform step S32; An adaptive smoothing unit, used to perform step S33, receives a control signal from the trick mark unit; A sequence generation unit is used to perform step S34.
9. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the vocal pitch detection method as described in any one of claims 1 to 5.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the vocal pitch detection method as described in any one of claims 1 to 5.