Oven voice recognition control method and related device based on artificial intelligence

Abstract

The present invention discloses an oven voice recognition control method based on artificial intelligence and a related device, and relates to the field of voice analysis technology. The method comprises: preprocessing voice command information; extracting features of the voice command information to obtain target acoustic features; performing language recognition based on the target acoustic features by using confidence analysis; performing semantic analysis on the voice command information based on the language category to obtain semantic text information; performing Mel smoothing coefficient calculation on the voice command information to extract voiceprint features, determining the voice output emotion according to the user's age group to perform response text matching; performing rhythm analysis on the target response text information, and performing rhythm adjustment on the initial response voice based on the rhythm analysis data; performing operation instruction matching based on the semantic text information, outputting the target response voice, and controlling the oven to execute the target operation instruction. The present invention enables the voice recognition control of the oven to meet the high intelligence requirements of users.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of speech analysis, and in particular to an oven speech recognition control method based on artificial intelligence and related devices. Background Art

[0002] Ovens are common cooking appliances in modern life. In order to improve the intelligence of oven control, various companies have gradually added voice control to the control function of ovens. At present, the recognition of different languages ​​is usually not taken into account in the voice control of ovens, resulting in low intelligence and applicability of oven voice control, and unable to meet the voice control needs of users of different languages. At the same time, the matching of response texts is usually directly matched according to the semantically parsed text information, and the voice output emotion is not considered to be added to the generation of the response text, resulting in the final output response voice being too stiff. In order to make the output response voice closer to the human voice, the response text is usually analyzed by rhythm. At present, it is usually achieved by the relevant personnel marking the response text by rhythm, but this method is too dependent on the professional quality of the relevant personnel and cannot guarantee the accuracy of the rhythm analysis of the response text, resulting in the target response voice failing to achieve the expected effect, so that the voice recognition control of the oven fails to meet the high intelligence needs of users. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art. The present invention provides an oven voice recognition control method and related devices based on artificial intelligence, which improves the reliability of oven voice recognition and enables the oven voice recognition control to meet the user's high intelligence needs.

[0004] In order to solve the above technical problems, the present invention provides an oven voice recognition control method based on artificial intelligence, which is applied to a cloud server and a voice interaction device of the oven, wherein the cloud server is communicatively connected with the voice interaction device, and the voice interaction device includes a voice acquisition unit and a voice response unit; the method includes:

[0005] Acquire the user's voice command information based on the voice acquisition unit, and transmit the voice command information to the cloud server, wherein the cloud server pre-processes the voice command information to obtain the pre-processed voice command information;

[0006] Perform feature extraction on the preprocessed voice command information to obtain target acoustic features;

[0007] Performing language recognition using confidence analysis based on the target acoustic features to obtain a language category corresponding to the preprocessed voice command information;

[0008] Performing semantic analysis on the preprocessed voice command information based on the language category to obtain corresponding semantic text information;

[0009] Calculating the Mel smoothing coefficient on the preprocessed voice command information to obtain the Mel smoothing coefficient, extracting the voiceprint feature based on the Mel smoothing coefficient to obtain the target voiceprint feature, determining the user age group based on the target voiceprint feature, determining the voice output emotion based on the user age group, matching the response text by language category based on the voice output emotion and semantic text information to obtain the target response text information;

[0010] Converting the target response text information into an initial response voice based on the voice output emotion, performing prosody analysis on the target response text information to obtain prosody analysis data, and performing prosody adjustment on the initial response voice based on the prosody analysis data to obtain a target response voice;

[0011] The operation instruction is matched based on the semantic text information to obtain the target operation instruction, the target response voice is transmitted to the voice response unit for output, and the oven is controlled to execute the target operation instruction.

[0012] Optionally, preprocessing the voice command information to obtain preprocessed voice command information includes:

[0013] Performing noise reduction processing on the voice command information to obtain the voice command information after noise reduction processing;

[0014] Performing frame processing on the voice command information after the noise reduction processing to obtain framed voice command information;

[0015] The framed voice command information is normalized to obtain pre-processed voice command information.

[0016] Optionally, the performing feature extraction on the preprocessed voice command information to obtain target acoustic features includes:

[0017] Performing discrete Fourier transform on the preprocessed voice command information to obtain a frequency domain signal;

[0018] Performing equal loudness pre-emphasis processing on the frequency domain signal to obtain a frequency domain signal after equal loudness pre-emphasis processing;

[0019] The first-order difference and the second-order difference are extracted from the frequency domain signal after equal-loudness pre-emphasis processing, and acoustic features are extracted based on the first-order difference and the second-order difference to obtain target acoustic features.

[0020] Optionally, performing language recognition by using confidence analysis based on the target acoustic feature to obtain a language category corresponding to the preprocessed voice command information includes:

[0021] Acquire language text segmentations and corresponding acoustic features of different languages ​​based on the language text corpus, and phonetically mark the language text segmentations based on preset language pronunciation rules to obtain corresponding language text segmentation phonemes;

[0022] Obtaining a first mapping relationship between language text segmentation and language text segmentation phonemes, and obtaining a second mapping relationship between acoustic features and language text segmentation, and constructing language acoustic models of different languages ​​based on the first mapping relationship and the second mapping relationship in combination with language knowledge graphs;

[0023] The target acoustic features are input into language acoustic models of different languages, and the language acoustic models of different languages ​​output corresponding confidences. The language corresponding to the language acoustic model with the largest confidence is selected as the language category corresponding to the preprocessed voice command information.

[0024] Optionally, performing semantic analysis on the preprocessed voice command information based on the language category to obtain corresponding semantic text information includes:

[0025] Determining a corresponding semantic parsing model based on the language category, and performing context information parsing on the preprocessed voice instruction information based on the semantic parsing model to obtain a context voice semantic set;

[0026] Based on the context speech semantic set combined with slot analysis, text generation is performed to obtain corresponding semantic text information.

[0027] Optionally, the step of calculating a Mel smoothing coefficient on the preprocessed voice command information to obtain a Mel smoothing coefficient, extracting voiceprint features based on the Mel smoothing coefficient to obtain target voiceprint features, determining a user age group based on the target voiceprint features, determining a voice output emotion based on the user age group, matching response texts by language category based on the voice output emotion and semantic text information to obtain target response text information includes:

[0028] Calculate the spectral line energy of the preprocessed voice command information to obtain the target spectral line energy;

[0029] Generate a spectrum envelope based on the target spectrum line energy, and perform discrete cosine transform cepstrum calculation based on the spectrum envelope to obtain a Mel smoothing coefficient;

[0030] Performing first-order difference extraction on the Mel smoothing coefficient to obtain the first-order difference of the Mel smoothing coefficient, and performing voiceprint feature extraction based on the Mel smoothing coefficient and the first-order difference of the Mel smoothing coefficient using a data fusion algorithm to obtain a target voiceprint feature;

[0031] Inputting the target voiceprint feature into a Gaussian distribution model to classify the age group, and obtaining the user age group;

[0032] Matching the corresponding voice output emotion in the emotion database based on the user's age group;

[0033] Performing semantic feature mapping processing on the semantic text information to obtain a semantic feature vector;

[0034] Obtaining a first label string of the semantic feature vector, obtaining a second label string of each initial response text in a response text library, and determining a joint probability distribution of the first label string and the second label string;

[0035] Calculating the similarity between the semantic feature vector and each initial response text based on the joint probability distribution, and selecting the initial response text with the highest similarity as the target initial response text;

[0036] Emotional feature words are determined based on the speech output emotion, and target response text information corresponding to the language category is generated based on the emotional feature words and the target initial response text.

[0037] Optionally, performing prosody analysis on the target response text information to obtain prosody analysis data, and performing prosody adjustment on the initial response voice based on the prosody analysis data to obtain the target response voice includes:

[0038] Performing prosodic structure division on the target response text information to obtain a target prosodic structure;

[0039] Acquire a historical response text and a historical response voice, extract a text prosodic feature of the historical response text and a voice prosodic feature of the historical response voice, and fuse the text prosodic feature and the voice prosodic feature to obtain a fused prosodic feature;

[0040] Determine the speech prosodic splicing point of the target response text information based on the fused prosodic feature combined with the target prosodic structure;

[0041] Generate prosodic analysis data based on the speech prosodic splicing points combined with prosodic sequence coding;

[0042] The initial response speech is rhythmically adjusted based on the rhythm analysis data combined with a speech splicing smoothing algorithm and a volume adjustment factor to obtain a target response speech.

[0043] In addition, the present invention also provides an oven voice recognition control device based on artificial intelligence, which is applied to a cloud server and a voice interaction device of the oven, wherein the cloud server is communicatively connected with the voice interaction device, and the voice interaction device includes a voice collection unit and a voice response unit; the device includes:

[0044] The voice command information preprocessing module is used to obtain the user's voice command information based on the voice acquisition unit, and transmit the voice command information to the cloud server, and the cloud server preprocesses the voice command information to obtain the preprocessed voice command information;

[0045] Feature extraction module: used to extract features from the preprocessed voice command information to obtain target acoustic features;

[0046] Language recognition module: used to perform language recognition based on the target acoustic features using confidence analysis to obtain the language category corresponding to the preprocessed voice command information;

[0047] Semantic parsing module: used to perform semantic parsing on the preprocessed voice command information based on the language category to obtain corresponding semantic text information;

[0048] Response text matching module: used to calculate the Mel smoothing coefficient of the preprocessed voice command information to obtain the Mel smoothing coefficient, and extract the voiceprint feature based on the Mel smoothing coefficient to obtain the target voiceprint feature, and determine the user age group based on the target voiceprint feature, determine the voice output emotion based on the user age group, and perform response text matching of language categories based on the voice output emotion and semantic text information to obtain the target response text information;

[0049] A speech prosody adjustment module: used for converting the target response text information into an initial response speech based on the speech output emotion, performing prosody analysis on the target response text information to obtain prosody analysis data, and performing prosody adjustment on the initial response speech based on the prosody analysis data to obtain the target response speech;

[0050] Voice and instruction output module: used to match operation instructions based on the semantic text information, obtain target operation instructions, transmit the target response voice to the voice response unit for output, and control the oven to execute the target operation instructions.

[0051] In addition, the present invention also provides an oven voice recognition control system based on artificial intelligence, the system includes a cloud server and a voice interaction device of the oven, the cloud server is communicatively connected to the voice interaction device, the voice interaction device includes a voice acquisition unit and a voice response unit; the system is configured to execute the above-mentioned oven voice recognition control method based on artificial intelligence.

[0052] In addition, the present invention also provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the above-mentioned artificial intelligence-based oven voice recognition control method.

[0053] In the embodiment of the present invention, language recognition is performed based on the target acoustic features using confidence analysis, which can more accurately identify different languages. The pre-processed voice instruction information is semantically parsed based on the language category, which can meet the voice semantic parsing of different languages ​​and improve the intelligent degree of oven voice control. The operation instruction matching is performed through the parsed semantic text information, which can improve the accuracy of operation instruction matching of users of different languages. Voiceprint feature extraction is performed based on the Mel smoothing coefficient to obtain the target voiceprint feature, and the user age group is determined based on the target voiceprint feature. The voice output emotion is determined based on the user age group, which can make the determined voice output emotion more reliable. The response text matching of the language category is performed based on the voice output emotion and semantic text information, which avoids the obtained response voice being too stiff and makes the output response voice more in line with the humanized setting. The target response text information is rhythmically analyzed, and the rhythm analysis data is generated by combining the speech rhythm splicing point with the rhythm sequence coding, and the initial response voice is rhythmically adjusted based on the rhythm analysis data to ensure the accuracy of the rhythm analysis and make the output response voice closer to the human voice. The target response voice is transmitted to the voice response unit for output, and the oven is controlled to execute the operation instructions. While the oven executes the relevant operation instructions, it can also output a more humanized response voice, so that the voice recognition control of the oven can meet the user's high intelligence needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 is a flow chart of an oven voice recognition control method based on artificial intelligence in an embodiment of the present invention;

[0056] Figure 2 Schematic diagram of the structure of an oven voice recognition control system based on artificial intelligence in an embodiment of the present invention;

[0057] Figure 3 It is a schematic diagram of the structural composition of an oven voice recognition control device based on artificial intelligence in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1

[0059] See also Figure 1 , Figure 1 : is a flow chart of an oven voice recognition control method based on artificial intelligence in an embodiment of the present invention, the method is applied to a cloud server and a voice interaction device of the oven, the cloud server is communicatively connected with the voice interaction device, the voice interaction device includes a voice acquisition unit and a voice response unit; the method includes:

[0060] S11: Acquire the user's voice command information based on the voice acquisition unit, and transmit the voice command information to the cloud server, where the cloud server pre-processes the voice command information to obtain pre-processed voice command information;

[0061] In the specific implementation process of the present invention, the preprocessing of the voice command information to obtain the preprocessed voice command information includes: performing noise reduction processing on the voice command information to obtain the voice command information after noise reduction processing; performing frame processing on the voice command information after noise reduction processing to obtain framed voice command information; performing normalization processing on the framed voice command information to obtain the preprocessed voice command information.

[0062] Specifically, based on the voice acquisition unit, the user's voice command information is obtained, and the voice command information can be an operation instruction for the oven, such as wake-up and baking time setting, and the voice command information is transmitted to the cloud server. The cloud server receives the voice command information, processes the voice command information, performs noise reduction processing on the voice command information, analyzes the silent part of the voice command information to obtain the silent part, estimates the background noise based on the silent part to obtain the background noise, performs spectrum subtraction on the background noise, and obtains the voice command information after the noise reduction processing. The voice command information after the noise reduction processing is framed, and the voice command information after the noise reduction processing is segmented according to the preset length to obtain the framed voice command information. The framed voice command information is normalized, and the framed voice command information is converged using a matrix for normalization processing. After the normalization processing is completed, the preprocessing of the voice command information is completed, and the preprocessed voice command information is obtained.

[0063] S12: extracting features from the preprocessed voice command information to obtain target acoustic features;

[0064] In the specific implementation process of the present invention, the feature extraction of the preprocessed voice command information to obtain the target acoustic feature includes: performing discrete Fourier transform on the preprocessed voice command information to obtain a frequency domain signal; performing equal loudness pre-emphasis processing on the frequency domain signal to obtain a frequency domain signal after equal loudness pre-emphasis processing; extracting first-order differences and second-order differences from the frequency domain signal after equal loudness pre-emphasis processing, and performing acoustic feature extraction based on the first-order differences and second-order differences to obtain the target acoustic feature.

[0065] Specifically, the pre-processed voice command information is subjected to a discrete Fourier transform, and the voice command information is transformed from the time domain to the frequency domain to obtain a frequency domain signal. The frequency domain signal is subjected to equal loudness pre-emphasis processing, and the frequencies in the frequency domain signal are mapped to a critical frequency band, which reflects the human hearing. The mapped frequencies are pre-emphasized using an equal loudness curve, which is a curve that reflects the relationship between voice loudness and frequency, to obtain a frequency domain signal subjected to equal loudness pre-emphasis processing. The first-order difference and the second-order difference are extracted from the frequency domain signal subjected to equal loudness pre-emphasis processing, and a first-order difference operation is performed on the frequency domain signal subjected to equal loudness pre-emphasis processing to obtain a first-order difference, and then a difference operation is performed on the first-order difference to obtain a second-order difference. Based on the first-order difference and the second-order difference, acoustic feature extraction is performed, Mel Frequency Cepstral Coefficients (MFCC) feature extraction is performed on the frequency domain signal after equal loudness pre-emphasis processing to obtain MFCC features, the first-order difference, the second-order difference and the MFCC features are combined to obtain multidimensional acoustic features, the multidimensional acoustic features are extended and optimized, and the multidimensional acoustic features are input into a feature extension optimization model to obtain extended and optimized multidimensional acoustic features, and the extended and optimized multidimensional acoustic features are differentially extended to obtain target acoustic features, so that the obtained acoustic features are more accurate, the acoustic features are highlighted and other interference information is suppressed.

[0066] S13: performing language recognition using confidence analysis based on the target acoustic features to obtain a language category corresponding to the preprocessed voice command information;

[0067] In the specific implementation process of the present invention, the language recognition is performed based on the target acoustic features using confidence analysis to obtain the language category corresponding to the preprocessed voice command information, including: obtaining language text segmentations and corresponding acoustic features of different languages ​​based on a language text corpus, and phonetically marking the language text segmentations based on preset language pronunciation rules to obtain corresponding language text segmentation phonemes; obtaining a first mapping relationship between language text segmentations and language text segmentation phonemes, and obtaining a second mapping relationship between acoustic features and language text segmentations, and constructing language acoustic models of different languages ​​based on the first mapping relationship and the second mapping relationship in combination with language knowledge graphs; inputting the target acoustic features into language acoustic models of different languages, the language acoustic models of different languages ​​output corresponding confidences, and selecting the language corresponding to the language acoustic model with the largest confidence as the language category corresponding to the preprocessed voice command information.

[0068] Specifically, based on the language text corpus, language text segmentation and corresponding acoustic features of different languages ​​are obtained, the languages ​​include different languages ​​such as Mandarin, English and German, and the language text segmentation is phonetically marked based on the preset language pronunciation rules, the preset language pronunciation rules contain pronunciation factor annotations of different languages, and the corresponding language text segmentation phonemes are obtained. A first mapping relationship between the language text segmentation and the language text segmentation phonemes is obtained, the association attributes of the language text segmentation and the language text segmentation factors are defined through the collection tag, the entity class attribute query of the language text segmentation and the language text segmentation factors is performed through the mapper function using the association attributes, the first entity class attribute of the language text segmentation and the second entity class attribute of the language text segmentation factor are obtained, the first mapping relationship is obtained by connection crawling according to the first entity class attribute and the second entity class attribute, and the second mapping relationship between the acoustic feature and the language text segmentation is obtained, and the acquisition step of the second mapping relationship is consistent with the acquisition step of the first mapping relationship. Based on the first mapping relationship and the second mapping relationship combined with the language knowledge graph, the language acoustic models of different languages ​​are constructed, and the word occurrence probability calculation model is constructed according to the first mapping relationship and the second mapping relationship. The word occurrence probability calculation model is combined with the language knowledge graph of different languages ​​to form the language acoustic models of different languages. The target acoustic features are input into the language acoustic models of different languages, and the language acoustic models of different languages ​​output corresponding confidences. Since the normalization calculation of the softmax function will cause the confidence probabilities calculated by the language acoustic models of different languages ​​to be inconsistent, a preset scaling factor is added to the confidence probability score items in any two compared language acoustic models of different languages, and the confidence output of the language acoustic model is scaled by the transformation of the preset scaling factor to obtain the final confidence, so that the confidence probabilities output by each model are consistent, ensuring the reliability of the confidence calculation, and comparing the confidences, and selecting the language corresponding to the language acoustic model with the largest confidence as the language category corresponding to the preprocessed voice instruction information.

[0069] S14: performing semantic analysis on the preprocessed voice command information based on the language category to obtain corresponding semantic text information;

[0070] In the specific implementation process of the present invention, the pre-processed voice command information is semantically parsed based on the language category to obtain corresponding semantic text information, including: determining a corresponding semantic parsing model based on the language category, and performing context information parsing on the pre-processed voice command information based on the semantic parsing model to obtain a contextual voice semantic set; text generation is performed based on the contextual voice semantic set combined with slot parsing to obtain corresponding semantic text information.

[0071] Specifically, different language categories correspond to different semantic parsing models, so the corresponding semantic parsing model needs to be determined according to the language category. Based on the semantic parsing model, the preprocessed voice command information is parsed for context information, and the preprocessed voice command information is processed by a dynamic programming algorithm to obtain the corresponding target state sequence. The state sequence processing can avoid repeated calculations and improve the computational efficiency of the model. The target state sequence is processed by a greedy algorithm to obtain a probability sequence. The beam search process can avoid falling into a local optimal solution and increase the probability of reaching a global optimal solution. The beam search is a process for probability sequence generation tasks. The probability sequence is input into the semantic parsing model to obtain a context speech semantic set. Text is generated based on the context speech semantic set combined with slot parsing, and the context speech semantic set is re-decoded by a number of decoders to obtain semantic re-decoded data, which can reduce the different semantics caused by the same factors in different language systems, and semantic conversion is performed on the semantic re-decoded data to obtain initial semantic text information, which is input into a semantic slot filling model to obtain corresponding slot values, and the corresponding slot values ​​are parsed to obtain key-value pair entity information, and user intent is constructed through the key-value pair entity information, and the initial semantic text information is combined with the derived user intent to obtain corresponding semantic text information, which satisfies speech semantic parsing in different languages ​​and makes the obtained semantic text information more consistent with the user's intended instructions, thereby being able to match more accurate operation instructions.

[0072] S15: Calculating the Mel smoothing coefficient of the preprocessed voice command information to obtain the Mel smoothing coefficient, extracting voiceprint features based on the Mel smoothing coefficient to obtain target voiceprint features, determining the user age group based on the target voiceprint features, determining the voice output emotion based on the user age group, matching the response texts by language category based on the voice output emotion and semantic text information to obtain target response text information;

[0073] In the specific implementation process of the present invention, the Mel smoothing coefficient is calculated on the preprocessed voice command information to obtain the Mel smoothing coefficient, and the voiceprint feature is extracted based on the Mel smoothing coefficient to obtain the target voiceprint feature, and the user age group is determined based on the target voiceprint feature, the voice output emotion is determined based on the user age group, and the response text of the language category is matched based on the voice output emotion and semantic text information to obtain the target response text information, including: calculating the spectral line energy of the preprocessed voice command information to obtain the target spectral line energy; generating a spectral envelope based on the target spectral line energy, and calculating the cepstrum of the discrete cosine transform based on the spectral envelope to obtain the Mel smoothing coefficient; performing first-order difference extraction on the Mel smoothing coefficient to obtain the first-order difference of the Mel smoothing coefficient, and using a data fusion algorithm based on the Mel smoothing coefficient and the first-order difference of the Mel smoothing coefficient. Perform voiceprint feature extraction to obtain target voiceprint features; input the target voiceprint features into a Gaussian distribution model for age group classification to obtain user age groups; match the corresponding voice output emotions in an emotion database based on the user age groups; perform semantic feature mapping on the semantic text information to obtain a semantic feature vector; obtain a first label string for the semantic feature vector, obtain a second label string for each initial response text in a response text library, and determine a joint probability distribution of the first label string and the second label string; calculate the similarity between the semantic feature vector and each initial response text based on the joint probability distribution, and select the initial response text with the highest similarity as the target initial response text; determine emotional feature words based on the voice output emotions, and generate target response text information corresponding to the language category based on the emotional feature words and the target initial response text.

[0074] Specifically, the spectral line energy is calculated for the preprocessed voice command information, the pitch period is estimated for the preprocessed voice command information, and a fast Fourier transform is performed based on the estimated pitch period to obtain the target spectral line energy. A spectral envelope is generated based on the target spectral line energy. The spectral envelope refers to a curve that sequentially connects the amplitude vertices of each frequency component in the sound spectrum to characterize the general trend of the spectrum. Based on the spectral envelope, a cepstrum of a discrete cosine transform is calculated, the spectral envelope is smoothed using a spectral envelope smoothing coefficient, and a Mel smoothing coefficient is calculated on the smoothed spectral envelope to obtain a Mel smoothing coefficient. The first-order difference extraction of the Mel smoothing coefficient is performed to obtain the first-order difference of the Mel smoothing coefficient, that is, the first-order difference operation is performed, and the voiceprint feature is extracted based on the Mel smoothing coefficient and the first-order difference of the Mel smoothing coefficient using a data fusion algorithm. The data fusion algorithm includes a mean subtraction method, a time series filtering method, and a weighted moving average filtering method. The Mel smoothing coefficient and the first-order difference of the Mel smoothing coefficient are combined to obtain the initial voiceprint feature. The initial voiceprint feature is mean subtracted according to the mean subtraction method, and the initial voiceprint feature after the mean subtraction is time series filtered according to the time series filtering method. The initial voiceprint feature after the time series filtering is adaptively averaged according to the weighted moving average filtering method. The noise effect is eliminated by the data fusion algorithm, so that the obtained voiceprint feature is closer to the actual situation, and the target voiceprint feature is obtained. The target voiceprint feature is input into the Gaussian distribution model for age group classification. The Gaussian distribution model contains several sub-Gaussian models, each of which corresponds to a different age group, and the user age group is obtained. Based on the user age group, the corresponding voice output emotion is matched in the emotion database, such as the user age group is 6-14 years old, the matched voice output emotion is happy and lively, etc., the user age group is 40-55 years old, and the matched voice output emotion is serious and firm, etc. The semantic text information is processed with semantic feature mapping to obtain a semantic feature vector. The first label string of the semantic feature vector is obtained, the second label string of each initial response text in the response text library is obtained, the joint probability distribution of the first label string and the second label string is determined, the normal distribution and exponential distribution of the first label string and the second label string are calculated through the naive Bayesian network, and the joint probability distribution is calculated according to the normal distribution and the exponential distribution. Based on the joint probability distribution, the similarity between the semantic feature vector and each initial response text is calculated, and a sampling string is generated according to the joint probability distribution. The first edit distance between the first label string and the sampling string is calculated, and the second edit distance between the second label string and the sampling string is calculated. The similarity is calculated according to the first edit distance and the second edit distance, and the initial response text with the highest similarity is selected as the target initial response text.Based on the voice output emotion, the emotional feature words are determined, different voice output emotions match different emotional words, and based on the emotional feature words and the target initial response text, the target response text information corresponding to the language category is generated, and the emotional words are added to the initial response text to generate the target response text, which corresponds to the language category. It is possible to make corresponding humanized responses according to the emotional features, thereby improving the user experience.

[0075] S16: converting the target response text information into an initial response voice based on the voice output emotion, performing prosody analysis on the target response text information to obtain prosody analysis data, and performing prosody adjustment on the initial response voice based on the prosody analysis data to obtain a target response voice;

[0076] In the specific implementation process of the present invention, the rhythm analysis of the target response text information is performed to obtain rhythm analysis data, and the rhythm adjustment of the initial response voice is performed based on the rhythm analysis data to obtain the target response voice, including: dividing the target response text information into rhythmic structures to obtain the target rhythmic structure; obtaining historical response texts and historical response voices, extracting text rhythmic features of the historical response texts and speech rhythmic features of the historical response voices, fusing the text rhythmic features and speech rhythmic features to obtain fused rhythmic features; determining the speech rhythmic splicing points of the target response text information based on the fused rhythmic features combined with the target rhythmic structure; generating rhythmic analysis data based on the speech rhythmic splicing points combined with rhythmic sequence coding; and rhythmically adjusting the initial response voice based on the rhythmic analysis data combined with a speech splicing smoothing algorithm and a volume adjustment factor to obtain the target response voice.

[0077] Specifically, the target response text information is converted into an initial response speech based on the speech output emotion, and the target response text information is converted into an initial response speech through speech conversion technology, and at the same time, the speech output emotion is used as the speech playback emotion and added to the generation of the initial response speech. The target response text information is divided into rhythmic structures, and the rhythmic structure of the target response text information is divided into rhythmic structures through a pre-trained rhythmic structure division model to obtain a target rhythmic structure, and the rhythmic structure includes rhythmic words, rhythmic phrases, and intonation phrases. The historical response text and the historical response speech are obtained, the text rhythmic features of the historical response text and the speech rhythmic features of the historical response speech are extracted, the text rhythmic features and the speech rhythmic features are fused, the speech rhythmic features are dimensionally transformed according to the dimensions of the text rhythmic features, and the speech rhythmic features after the dimension transformation are obtained, the target matrix of the text rhythmic features and the speech rhythmic features after the dimension transformation is obtained, the target matrix is ​​subjected to singular value decomposition, a feature mapping matrix is ​​obtained, and the text rhythmic features and the speech rhythmic features after the dimension transformation are fused according to the feature mapping matrix to obtain a fused rhythmic feature. Based on the fused prosodic features combined with the target prosodic structure, the speech prosodic splicing points of the target response text information are determined. According to the fused prosodic features and the target prosodic structure, the speech prosodic splicing points of the target response text information are determined by using the speech splicing cost analysis. The speech splicing cost analysis is implemented by using a preset splicing cost analysis function, which can improve the accuracy of the speech prosodic splicing points. Based on the speech prosodic splicing points combined with the prosodic sequence coding, prosodic analysis data is generated. According to the speech prosodic splicing points, embedded information of prosodic words and prosodic phrases is formed. According to the embedded information of prosodic words and prosodic phrases, prosodic sequence coding is generated using an attention mechanism. Based on the prosodic analysis data combined with the speech splicing smoothing algorithm and the volume adjustment factor, the initial response speech is prosodic adjusted. According to the prosodic analysis data, the initial response speech is preliminarily adjusted in prosodic manner to obtain the initial response speech after the preliminarily adjusted prosodic manner. The initial response speech after the preliminarily adjusted prosodic manner is smoothed by using the speech splicing smoothing algorithm. The volume of the response speech is determined according to the volume adjustment factor, and finally the target response speech of the corresponding language is obtained. The target response voice is made smoother and more fluent, and the naturalness and expressiveness of the voice are improved.

[0078] S17: Match the operation instructions based on the semantic text information to obtain the target operation instruction, transmit the target response voice to the voice response unit for output, and control the oven to execute the target operation instruction.

[0079] In the specific implementation process of the present invention, operation instruction matching is performed based on the semantic text information to obtain the target operation instruction. The user's instruction intention can be learned based on the semantic text information obtained by semantically parsing the voice instruction information issued by the user. The corresponding target operation instruction is matched in the instruction library based on the semantic text information to meet the operation instruction matching of users of different languages. The target operation instruction is transmitted to the oven, and the oven executes the target operation instruction. At the same time, the target response voice is transmitted to the voice response unit for output. The oven executes the corresponding operation instruction while outputting the target response voice, realizing the high intelligence of the oven voice recognition control and meeting the user's high requirements for oven voice control.

[0080] In the embodiment of the present invention, language recognition is performed based on the target acoustic features using confidence analysis, which can more accurately identify different languages. The pre-processed voice instruction information is semantically parsed based on the language category, which can meet the voice semantic parsing of different languages ​​and improve the intelligent degree of oven voice control. The operation instruction matching is performed through the parsed semantic text information, which can improve the accuracy of operation instruction matching of users of different languages. Voiceprint feature extraction is performed based on the Mel smoothing coefficient to obtain the target voiceprint feature, and the user age group is determined based on the target voiceprint feature. The voice output emotion is determined based on the user age group, which can make the determined voice output emotion more reliable. The response text matching of the language category is performed based on the voice output emotion and semantic text information, which avoids the obtained response voice being too stiff and makes the output response voice more in line with the humanized setting. The target response text information is rhythmically analyzed, and the rhythm analysis data is generated by combining the speech rhythm splicing point with the rhythm sequence coding, and the initial response voice is rhythmically adjusted based on the rhythm analysis data to ensure the accuracy of the rhythm analysis and make the output response voice closer to the human voice. The target response voice is transmitted to the voice response unit for output, and the oven is controlled to execute the operation instructions. While the oven executes the relevant operation instructions, it can also output a more humanized response voice, so that the voice recognition control of the oven can meet the user's high intelligence needs. Embodiment 2

[0081] See also Figure 2 , Figure 2 2 is a schematic diagram of the structural composition of an oven voice recognition control system based on artificial intelligence in an embodiment of the present invention. The system includes a cloud server 21 and a voice interaction device 22 of the oven. The cloud server 21 is communicatively connected with the voice interaction device 22. The voice interaction device 22 includes a voice acquisition unit and a voice response unit. The system is configured to execute the oven voice recognition control method based on artificial intelligence in the above embodiment.

[0082] In the specific implementation process of the present invention, the cloud server 21 is used to analyze the acquired voice command information to match the corresponding operation instructions and response voices. The voice interaction device 22 is communicated with the cloud server 21 to realize artificial intelligence-based voice recognition control of the oven. The voice collection unit in the voice interaction device 22 is used to collect the user's voice command information, and the voice response unit is used to play the response voice.

[0083] at the same time, Figure 2 The illustrated artificial intelligence-based oven voice recognition control system does not constitute a limitation on all components, and may include more or fewer components than illustrated, or combine certain components. The specific implementation method can be found in the above embodiments, which will not be described in detail here.

[0084] In the embodiment of the present invention, language recognition is performed based on the target acoustic features using confidence analysis, which can more accurately identify different languages. The pre-processed voice instruction information is semantically parsed based on the language category, which can meet the voice semantic parsing of different languages ​​and improve the intelligent degree of oven voice control. The operation instruction matching is performed through the parsed semantic text information, which can improve the accuracy of operation instruction matching of users of different languages. Voiceprint feature extraction is performed based on the Mel smoothing coefficient to obtain the target voiceprint feature, and the user age group is determined based on the target voiceprint feature. The voice output emotion is determined based on the user age group, which can make the determined voice output emotion more reliable. The response text matching of the language category is performed based on the voice output emotion and semantic text information, which avoids the obtained response voice being too stiff and makes the output response voice more in line with the humanized setting. The target response text information is rhythmically analyzed, and the rhythm analysis data is generated by combining the speech rhythm splicing point with the rhythm sequence coding, and the initial response voice is rhythmically adjusted based on the rhythm analysis data to ensure the accuracy of the rhythm analysis and make the output response voice closer to the human voice. The target response voice is transmitted to the voice response unit for output, and the oven is controlled to execute the operation instructions. While the oven executes the relevant operation instructions, it can also output a more humanized response voice, so that the voice recognition control of the oven can meet the user's high intelligence needs. Embodiment 3

[0085] See also Figure 3 , Figure 3 : is a structural schematic diagram of an oven voice recognition control device based on artificial intelligence in an embodiment of the present invention, the device is applied to a cloud server and a voice interaction device of the oven, the cloud server is communicatively connected with the voice interaction device, the voice interaction device includes a voice collection unit and a voice response unit; the device includes:

[0086] The voice command information preprocessing module 31 is used to obtain the user's voice command information based on the voice acquisition unit, and transmit the voice command information to the cloud server, and the cloud server preprocesses the voice command information to obtain the preprocessed voice command information;

[0087] Feature extraction module 32: used to extract features from the pre-processed voice command information to obtain target acoustic features;

[0088] Language identification module 33: used to perform language identification based on the target acoustic features using confidence analysis to obtain the language category corresponding to the pre-processed voice instruction information;

[0089] Semantic parsing module 34: used to perform semantic parsing on the pre-processed voice instruction information based on the language category to obtain corresponding semantic text information;

[0090] The response text matching module 35 is used to calculate the Mel smoothing coefficient of the preprocessed voice instruction information to obtain the Mel smoothing coefficient, and extract the voiceprint feature based on the Mel smoothing coefficient to obtain the target voiceprint feature, and determine the user age group based on the target voiceprint feature, determine the voice output emotion based on the user age group, and perform response text matching of language categories based on the voice output emotion and semantic text information to obtain the target response text information;

[0091] A speech prosody adjustment module 36 is used to convert the target response text information into an initial response speech based on the speech output emotion, perform prosody analysis on the target response text information to obtain prosody analysis data, and perform prosody adjustment on the initial response speech based on the prosody analysis data to obtain the target response speech;

[0092] Voice and instruction output module 37: used to match operation instructions based on the semantic text information, obtain target operation instructions, transmit the target response voice to the voice response unit for output, and control the oven to execute the target operation instructions.

[0093] In the specific implementation process of the present invention, the specific implementation methods of the device items can be referred to the above embodiments, which will not be repeated here.

[0094] In the embodiment of the present invention, language recognition is performed based on the target acoustic features using confidence analysis, which can more accurately identify different languages. The pre-processed voice instruction information is semantically parsed based on the language category, which can meet the voice semantic parsing of different languages ​​and improve the intelligent degree of oven voice control. The operation instruction matching is performed through the parsed semantic text information, which can improve the accuracy of operation instruction matching of users of different languages. Voiceprint feature extraction is performed based on the Mel smoothing coefficient to obtain the target voiceprint feature, and the user age group is determined based on the target voiceprint feature. The voice output emotion is determined based on the user age group, which can make the determined voice output emotion more reliable. The response text matching of the language category is performed based on the voice output emotion and semantic text information, which avoids the obtained response voice being too stiff and makes the output response voice more in line with the humanized setting. The target response text information is rhythmically analyzed, and the rhythm analysis data is generated by combining the speech rhythm splicing point with the rhythm sequence coding, and the initial response voice is rhythmically adjusted based on the rhythm analysis data to ensure the accuracy of the rhythm analysis and make the output response voice closer to the human voice. The target response voice is transmitted to the voice response unit for output, and the oven is controlled to execute the operation instructions. While the oven executes the relevant operation instructions, it can also output a more humanized response voice, so that the voice recognition control of the oven can meet the user's high intelligence needs.

[0095] A computer-readable storage medium is provided in an embodiment of the present invention, wherein a computer program is stored on the computer-readable storage medium, and when the program is executed by a processor, an oven voice recognition control method based on artificial intelligence is implemented in any one of the above embodiments. The computer-readable storage medium includes, but is not limited to, any type of disk (including floppy disk, hard disk, optical disk, CD-ROM, and magneto-optical disk), ROM (Read-Only Memory), RAM (Random Access Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, magnetic card or optical card. That is, the storage device includes any medium that can store or transmit information in a readable form by a device (for example, a computer, a mobile phone), which can be a read-only memory, a disk or an optical disk, etc.

[0096] In addition, the above is a detailed introduction to an oven voice recognition control method based on artificial intelligence and related devices provided in an embodiment of the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for a general technician in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. An oven voice recognition control method based on artificial intelligence, characterized in that: A voice interaction device applied to a cloud server and an oven, wherein the cloud server is communicatively connected with the voice interaction device, and the voice interaction device includes a voice collection unit and a voice response unit; the method includes: Acquire the user's voice command information based on the voice acquisition unit, and transmit the voice command information to the cloud server, wherein the cloud server pre-processes the voice command information to obtain the pre-processed voice command information; Perform feature extraction on the preprocessed voice command information to obtain target acoustic features; Performing language recognition using confidence analysis based on the target acoustic features to obtain a language category corresponding to the preprocessed voice command information; Performing semantic analysis on the preprocessed voice command information based on the language category to obtain corresponding semantic text information; Calculating the Mel smoothing coefficient on the preprocessed voice command information to obtain the Mel smoothing coefficient, extracting the voiceprint feature based on the Mel smoothing coefficient to obtain the target voiceprint feature, determining the user age group based on the target voiceprint feature, determining the voice output emotion based on the user age group, matching the response text by language category based on the voice output emotion and semantic text information to obtain the target response text information; Based on the speech output emotion, the target response text information is converted into an initial response speech, the target response text information is subjected to rhythmic analysis to obtain rhythmic analysis data, and the initial response speech is rhythmically adjusted based on the rhythmic analysis data to obtain the target response speech, including: performing rhythmic structure division on the target response text information to obtain a target rhythmic structure; acquiring historical response texts and historical response speech, extracting text rhythmic features of the historical response texts and speech rhythmic features of the historical response speech, fusing the text rhythmic features and speech rhythmic features to obtain fused rhythmic features; determining the speech rhythmic splicing points of the target response text information based on the fused rhythmic features combined with the target rhythmic structure; generating rhythmic analysis data based on the speech rhythmic splicing points combined with rhythmic sequence coding; rhythmically adjusting the initial response speech based on the rhythmic analysis data combined with a speech splicing smoothing algorithm and a volume adjustment factor to obtain a target response speech; The operation instruction is matched based on the semantic text information to obtain the target operation instruction, the target response voice is transmitted to the voice response unit for output, and the oven is controlled to execute the target operation instruction.

2. The oven voice recognition control method based on artificial intelligence according to claim 1, characterized in that: The preprocessing of the voice command information to obtain the preprocessed voice command information includes: Performing noise reduction processing on the voice command information to obtain the voice command information after noise reduction processing; Performing frame processing on the voice command information after the noise reduction processing to obtain framed voice command information; The framed voice command information is normalized to obtain pre-processed voice command information.

3. The oven voice recognition control method based on artificial intelligence according to claim 1, characterized in that: The step of extracting features from the preprocessed voice command information to obtain target acoustic features includes: Performing discrete Fourier transform on the preprocessed voice command information to obtain a frequency domain signal; Performing equal loudness pre-emphasis processing on the frequency domain signal to obtain a frequency domain signal after equal loudness pre-emphasis processing; The first-order difference and the second-order difference are extracted from the frequency domain signal after equal-loudness pre-emphasis processing, and acoustic features are extracted based on the first-order difference and the second-order difference to obtain target acoustic features.

4. The oven voice recognition control method based on artificial intelligence according to claim 1, characterized in that: The performing language recognition based on the target acoustic features by using confidence analysis to obtain the language category corresponding to the preprocessed voice instruction information includes: Acquire language text segmentations and corresponding acoustic features of different languages ​​based on the language text corpus, and phonetically mark the language text segmentations based on preset language pronunciation rules to obtain corresponding language text segmentation phonemes; Obtaining a first mapping relationship between language text segmentation and language text segmentation phonemes, and obtaining a second mapping relationship between acoustic features and language text segmentation, and constructing language acoustic models of different languages ​​based on the first mapping relationship and the second mapping relationship in combination with language knowledge graphs; The target acoustic features are input into language acoustic models of different languages, and the language acoustic models of different languages ​​output corresponding confidences. The language corresponding to the language acoustic model with the largest confidence is selected as the language category corresponding to the preprocessed voice command information.

5. The oven voice recognition control method based on artificial intelligence according to claim 1, characterized in that: The performing semantic analysis on the preprocessed voice command information based on the language category to obtain corresponding semantic text information includes: Determining a corresponding semantic parsing model based on the language category, and performing context information parsing on the preprocessed voice instruction information based on the semantic parsing model to obtain a context voice semantic set; Based on the context speech semantic set combined with slot analysis, text generation is performed to obtain corresponding semantic text information.

6. The oven voice recognition control method based on artificial intelligence according to claim 1, characterized in that: The method further comprises: calculating the Mel smoothing coefficient of the preprocessed voice command information to obtain the Mel smoothing coefficient, extracting the voiceprint feature based on the Mel smoothing coefficient to obtain the target voiceprint feature, determining the user age group based on the target voiceprint feature, determining the voice output emotion based on the user age group, matching the response text by language category based on the voice output emotion and semantic text information to obtain the target response text information, including: Calculate the spectral line energy of the preprocessed voice command information to obtain the target spectral line energy; Generate a spectrum envelope based on the target spectrum line energy, and perform discrete cosine transform cepstrum calculation based on the spectrum envelope to obtain a Mel smoothing coefficient; Performing first-order difference extraction on the Mel smoothing coefficient to obtain the first-order difference of the Mel smoothing coefficient, and performing voiceprint feature extraction based on the Mel smoothing coefficient and the first-order difference of the Mel smoothing coefficient using a data fusion algorithm to obtain a target voiceprint feature; Inputting the target voiceprint feature into a Gaussian distribution model to classify the age group, and obtaining the user age group; Matching the corresponding voice output emotion in the emotion database based on the user's age group; Performing semantic feature mapping processing on the semantic text information to obtain a semantic feature vector; Obtaining a first label string of the semantic feature vector, obtaining a second label string of each initial response text in a response text library, and determining a joint probability distribution of the first label string and the second label string; Calculating the similarity between the semantic feature vector and each initial response text based on the joint probability distribution, and selecting the initial response text with the highest similarity as the target initial response text; Emotional feature words are determined based on the speech output emotion, and target response text information corresponding to the language category is generated based on the emotional feature words and the target initial response text.

7. An oven voice recognition control device based on artificial intelligence, characterized in that: A voice interaction device applied to a cloud server and an oven, wherein the cloud server is communicatively connected with the voice interaction device, and the voice interaction device includes a voice collection unit and a voice response unit; the device includes: The voice command information preprocessing module is used to obtain the user's voice command information based on the voice acquisition unit, and transmit the voice command information to the cloud server, and the cloud server preprocesses the voice command information to obtain the preprocessed voice command information; Feature extraction module: used to extract features from the preprocessed voice command information to obtain target acoustic features; Language recognition module: used to perform language recognition based on the target acoustic features using confidence analysis to obtain the language category corresponding to the preprocessed voice command information; Semantic parsing module: used to perform semantic parsing on the preprocessed voice command information based on the language category to obtain corresponding semantic text information; Response text matching module: used to calculate the Mel smoothing coefficient of the preprocessed voice command information to obtain the Mel smoothing coefficient, and extract the voiceprint feature based on the Mel smoothing coefficient to obtain the target voiceprint feature, and determine the user age group based on the target voiceprint feature, determine the voice output emotion based on the user age group, and perform response text matching of language categories based on the voice output emotion and semantic text information to obtain the target response text information; A speech prosody adjustment module: used to convert the target response text information into an initial response speech based on the speech output emotion, perform prosody analysis on the target response text information to obtain prosody analysis data, and perform prosody adjustment on the initial response speech based on the prosody analysis data to obtain the target response speech, including: performing prosody structure division on the target response text information to obtain a target prosody structure; obtaining historical response texts and historical response speech, extracting text prosody features of the historical response texts and speech prosody features of the historical response speech, fusing the text prosody features and speech prosody features to obtain fused prosody features; determining the speech prosody splicing point of the target response text information based on the fused prosody features combined with the target prosody structure; generating prosody analysis data based on the speech prosody splicing point combined with prosody sequence coding; performing prosody adjustment on the initial response speech based on the prosody analysis data combined with a speech splicing smoothing algorithm and a volume adjustment factor to obtain a target response speech; Voice and instruction output module: used to match operation instructions based on the semantic text information, obtain target operation instructions, transmit the target response voice to the voice response unit for output, and control the oven to execute the target operation instructions.

8. An oven voice recognition control system based on artificial intelligence, characterized in that: The system includes a cloud server and a voice interaction device of an oven, the cloud server is communicatively connected to the voice interaction device, and the voice interaction device includes a voice acquisition unit and a voice response unit; the system is configured to execute the artificial intelligence-based oven voice recognition control method described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed on an electronic device, the electronic device executes the artificial intelligence-based oven voice recognition control method as described in any one of claims 1 to 6.

Images