A speech translation method and device for resolving speech ambiguity

Abstract

The invention relates to the technical field of speech translation data processing, and specifically discloses a speech translation method and device for solving speech ambiguity. The method comprises: acquiring speech data, analyzing the speech data, and adjusting translation speech feature data. The invention can effectively remove background noise in the speech data, reduce environmental interference, improve speech clarity and recognizability, and provide high-quality speech data support for subsequent speech recognition and translation by performing preprocessing operations on speech data to be translated. Preprocessing and adjusting translation speech feature data can effectively identify and eliminate ambiguity in the speech data, ensure that the translation result is more accurate, and improve the efficiency of translation. Finally, speech ambiguity is fed back through a speech translation feedback device, so that ambiguity and errors in the translation process can be timely identified and adjusted, and the accuracy of speech translation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of speech translation data processing, and in particular to a speech translation method and device for resolving speech ambiguity. Background Art

[0002] At present, with the increasing demand for speech translation, speech translation technology has become an important bridge connecting different languages ​​and cultures. However, in practical applications, speech translation faces many challenges, among which speech ambiguity is particularly prominent. When translating speech data, existing speech translation technologies often rely on pre-built dictionaries and models, and have limited recognition and processing capabilities for new words not included in the dictionary or speech data in special contexts, which may lead to mistranslation or omission in the translation process, affecting the accuracy of the translation.

[0003] For example, the invention patent with the announcement number CN117094329B announces a speech translation method and device for resolving speech ambiguity, which relates to the field of speech translation technology. It includes: obtaining speech data to be translated; constructing a homonym dictionary; inputting the speech data into the constructed speech translation model; and obtaining the translation text of the speech data according to the speech data, the homonym dictionary and the speech translation model. An efficient speech disambiguation method is constructed, which can effectively alleviate the ambiguous phenomenon in the speech translation model and improve the accuracy of speech translation.

[0004] For example, the invention patent with the announcement number CN111310481B announces a speech translation method, device, computer equipment and storage medium, which relates to the speech technology field of computer technology. The specific implementation scheme is: by obtaining the i-th string of the source speech, where i is a positive integer; inputting the i-th string into the trained segmentation model to determine whether the i-th string is an unambiguous semantic unit; if the i-th string is determined to be an unambiguous semantic unit, the i-th string is translated to generate the i-th target string; and if the i-th string is determined not to be an unambiguous semantic unit, the i-th string is not translated until the sum of the i-th string to the i+nth string is an unambiguous semantic unit, and the sum of the i-th string to the i+nth string is translated, where n is a positive integer. In this way, the problem of string translation errors with multiple interpretations in the source speech, which affects the accuracy of the whole sentence translation, is avoided, which is conducive to improving the accuracy of speech translation.

[0005] Based on the above solution, it is found that the collected voice data needs to be analyzed and translated. However, in the current technical solutions for processing voice data, there are limitations in the analysis of voice data, which may lead to mistranslation, omission or semantic deviation when processing voice data. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention provides a speech translation method and device for resolving speech ambiguity, which can effectively solve the problems involved in the above-mentioned background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: In a first aspect, the present invention provides a speech translation method for resolving speech ambiguity, comprising: S1. Acquiring speech data: a data processing device acquires speech data to be translated through a recording device, and performs preprocessing operations on the speech data to be translated, and records the preprocessed speech data to be translated as speech data to be translated; S2. Speech data analysis: a speech translation device performs initial speech translation on the speech data to be translated, and a data processing device extracts the translation result corresponding to the initial speech translation to obtain first translation result feature data and translated speech feature data, and performs a comprehensive analysis to obtain a speech translation quality index; S3. Translation speech ambiguity adjustment: the speech translation quality index is compared with a speech translation quality index boundary value preset in a speech translation database, and if the speech translation quality index is less than the speech translation quality index preset in the speech translation database, the speech translation quality index is adjusted. According to the indicator definition value, a preprocessing operation is performed on the speech feature data to be translated, and the preprocessed speech feature data to be translated is recorded as adjusted speech data, whereby the speech translation device performs an ambiguous translation adjustment operation on the adjusted speech data, and records the speech data after the ambiguous translation adjustment operation is completed as adjusted speech data, based on the ambiguous translation adjustment process, feature data of the adjusted speech is extracted, and a speech translation adjustment index is obtained through analysis, and the index is compared with a preset speech translation adjustment index threshold, based on the comparison result, a second translation is performed on the adjusted speech data after the ambiguous translation adjustment is completed, and the second translation result data is obtained, and the second translation result data is comprehensively analyzed with the speech translation adjustment index to obtain an ambiguous translation adjustment accuracy index, and the ambiguous translation adjustment accuracy index is compared with the ambiguous translation adjustment accuracy index threshold preset in the speech translation database, and finally the speech ambiguous translation adjustment process is fed back through a speech translation feedback device.

[0008] As a further method, the speech translation quality index, the specific analysis process is:

[0009] Speech translation quality indicators, the specific analysis process is as follows:

[0010] The amplitude of each frequency component of the speech signal is squared and the obtained value is recorded as the power of each frequency component. The values ​​are added together to obtain the total power of the frequency components. The total number of frequency components is counted, and the total power of the frequency components is divided by the total number of frequency components to finally obtain the average power of the speech signal.

[0011] The first speech translation rate, speech translation duration, amplitude, average power and average frequency of formant of speech signal are comprehensively analyzed to obtain the speech translation quality index.

[0012] As a further method, the adjusting of the characteristic data of the speech specifically includes adjusting the average power, the average frequency of the formant and the average bandwidth of the speech signal, and adjusting the clarity data of the speech signal.

[0013] The adjusting the clarity data of the speech signal specifically includes adjusting the amplitude and fundamental frequency of the speech signal and adjusting the speech speed of the speech.

[0014] As a further method, the speech translation adjustment index, the specific analysis process is:

[0015] The amplitude and fundamental frequency of the adjusted speech signal and the speaking speed of the adjusted speech are comprehensively analyzed to obtain the clarity index of the adjusted speech.

[0016] The average power, average frequency of formant, average bandwidth and clarity index of the adjusted speech signal are comprehensively analyzed to obtain the speech translation adjustment index. The specific analysis formula is:

[0017]

[0018] In the formula, is the index for adjusting the speech translation, e is a natural constant, To adjust the average power of the speech signal, The preset reference power of the adjusted speech signal for the speech translation database, The correction factor corresponding to the average power of the adjusted speech signal preset in the speech translation database, To adjust the average frequency of the formant of the speech signal, Adjust the speech signal formant reference frequency preset for the speech translation database, The correction factor corresponding to the average frequency of the formant of the speech signal preset in the speech translation database, To adjust the average bandwidth of the speech signal, The impact factor corresponding to the bandwidth unit value of the adjusted speech signal preset in the speech translation database, To adjust the speech clarity index, The weight factor corresponding to the speech clarity index preset in the speech translation database is adjusted.

[0019] As a further method, feedback is provided to the speech ambiguous translation adjustment process, and the specific feedback process is:

[0020] After the ambiguous translation adjustment is completed, the adjusted speech data is translated for the second time to obtain second translation result data, and a comprehensive analysis is performed with the speech translation adjustment index to obtain the ambiguous translation adjustment accuracy index.

[0021] The ambiguous translation adjustment accuracy index is compared with the ambiguous translation adjustment accuracy index threshold preset in the speech translation database. If the ambiguous translation adjustment accuracy index is greater than or equal to the ambiguous translation adjustment accuracy index threshold preset in the speech translation database, there is no need to provide feedback on the speech ambiguous translation adjustment process through the speech translation feedback device.

[0022] If the ambiguous translation adjustment accuracy index is less than the ambiguous translation adjustment accuracy index threshold preset in the speech translation database, the speech ambiguous translation adjustment process is fed back through the speech translation feedback device.

[0023] A second aspect of the present invention provides a device for a speech translation method for resolving speech ambiguity, comprising: a recording device, a data processing device, a speech translation device, and a speech translation feedback device.

[0024] The recording device is used to obtain voice data that needs to be translated.

[0025] The data processing device is used to receive voice data to be translated through the Internet of Things, and perform preprocessing to obtain voice data to be translated, receive and preprocess voice ambiguous data in the initial translation voice data of the voice translation device, obtain adjusted voice data, respectively analyze the first translation result feature data and the feature data of the adjusted voice, and finally obtain a voice translation quality index and a voice translation adjustment index.

[0026] The speech translation device is used to receive the speech data to be translated from the data processing device through the Internet of Things and translate it to obtain the initial translated speech data, and receive and adjust the adjusted speech data of the data processing device.

[0027] The speech translation feedback device is used to receive the speech translation quality index and the speech translation adjustment index of the data processing device through the Internet of Things, and finally provide feedback on the speech ambiguous translation adjustment process.

[0028] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0029] (1) By preprocessing the speech data to be translated, the present invention can effectively remove background noise in the speech data, reduce environmental interference, improve the clarity and recognizability of the speech, and provide high-quality speech data support for subsequent speech recognition and translation.

[0030] (2) The present invention can effectively identify and eliminate ambiguity in speech data by preprocessing the translated speech feature data and adjusting ambiguity, thereby ensuring that the translation result is more accurate and improving the efficiency of translation.

[0031] (3) The present invention can identify common error patterns, fuzzy areas or specific language phenomena in the translation process by analyzing the first translation result data and the feature data of the speech to be translated, and then analyze to obtain a speech translation quality index. The speech translation quality index can serve as a basis for subsequent translation adjustment and optimization to help improve the accuracy of the overall translation. At the same time, the present invention can more accurately identify and solve fuzzy and uncertain problems in translation by performing special translation adjustments on ambiguous data and extracting feature data in the adjustment process. This refined processing helps to improve the accuracy and reliability of the translation results.

[0032] (4) The present invention provides feedback on speech ambiguity through a speech translation feedback device, which can timely identify and adjust ambiguities and errors in the translation process, thereby improving the accuracy of speech translation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0034] Figure 1 The figure is a schematic flow chart of the method steps of the present invention.

[0035] Figure 2 This is a simplified connection diagram of the equipment involved in the present invention.

[0036] Figure 3 It is a simulation curve diagram of the average bandwidth and the average frequency of the resonance peak involved in the present invention. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0038] Reference Figure 1As shown, the first aspect of the present invention provides a speech translation method for resolving speech ambiguity, comprising: S1. Acquiring speech data: a data processing device acquires speech data to be translated through a recording device, and performs a preprocessing operation on the speech data to be translated, and records the preprocessed speech data to be translated as speech data to be translated.

[0039] S2. Speech data analysis: The speech translation device performs initial speech translation on the speech data to be translated, and the data processing device extracts the translation result corresponding to the initial speech translation to obtain first translation result data and speech feature data to be translated, and performs comprehensive analysis to obtain speech translation quality indicators.

[0040] S3. Translation speech ambiguity adjustment: compare the speech translation quality index with the speech translation quality index boundary value preset in the speech translation database. If the speech translation quality index is less than the speech translation quality index boundary value preset in the speech translation database, perform a preprocessing operation on the speech feature data to be translated, record the preprocessed speech feature data to be translated as adjusted speech data, thereby the speech translation device performs an ambiguous translation adjustment operation on the adjusted speech data, and records the speech data after the ambiguous translation adjustment operation is completed as adjusted speech data. Based on the ambiguous translation adjustment process, extract the feature data of the adjusted speech, obtain the speech translation adjustment index through analysis, and compare it with the preset speech translation adjustment index threshold. Based on the comparison result, perform a second translation on the adjusted speech data after the ambiguous translation adjustment is completed to obtain the second translation result data, and perform a comprehensive analysis with the speech translation adjustment index to obtain the ambiguous translation adjustment accuracy index. Compare the ambiguous translation adjustment accuracy index with the ambiguous translation adjustment accuracy index threshold preset in the speech translation database, and finally provide feedback on the speech ambiguous translation adjustment process through the speech translation feedback device.

[0041] In a specific embodiment, speech ambiguity is particularly common during the process of on-site real-time speech translation. For example, when the speaker uses complex or ambiguous terms, initial translation may lead to misunderstandings. Due to cultural and contextual differences between different languages, speech ambiguity may lead to inaccurate translation results and affect communication effectiveness. Because the context and subtle differences between multiple languages ​​need to be considered during the translation process, it is difficult to directly adjust speech ambiguity. Therefore, the present invention performs fine adjustments through a data processing device and a speech translation device: the data processing device is responsible for preprocessing and analysis of speech data; the speech translation device is a dedicated translation terminal for processing and adjusting translation data. Through the cooperation of these devices, the accuracy and reliability of the final translation result are ensured.

[0042] Specifically, the preprocessing operation on the speech data to be translated is to monitor the noise of each speech data point in the speech data to be translated to obtain the noise value of each speech data point. If the noise value of a speech data point is higher than the noise reference threshold preset in the speech translation database, the speech data point is subjected to noise reduction processing. Thus, the statistics of each speech data point on which noise reduction has been completed and each speech data point on which noise reduction has not been completed are counted, which is the speech data to be translated that has been finally preprocessed and is recorded as the speech data to be translated.

[0043] It should be explained that the above-mentioned speech data points are multiple data of speech data that need to be translated, which are recorded as speech data points in this embodiment. The preset noise reference threshold refers to the value used to determine whether a certain speech data point needs to be subjected to noise reduction processing during the process of reducing the noise of speech data. The noise reference threshold can be determined based on previous historical experience and historical experimental data. When the data noise value of a certain speech data point is higher than the noise reference threshold preset in the speech translation database, it means that the data point is subject to relatively significant noise pollution, and its quality is not enough to be directly used for speech translation processing or may seriously affect the accuracy of the translation result. Therefore, it is necessary to perform noise reduction processing on these high-noise data points. The specific noise reduction processing can be performed by the spectrum subtraction noise reduction method in the data processing device, which subtracts the spectrum of the noise from the spectrum of the noisy speech signal to obtain a relatively pure speech signal, improve its quality and reduce the impact of noise on the translation result. The noise value of each speech data point refers to the noise intensity of each speech data point, in decibels. In this embodiment, if the noise value of a certain speech data point is lower than or equal to the noise reference threshold preset in the speech translation database, there is no need to perform noise reduction processing on the speech data point, and the speech data is directly recorded as the speech data to be translated.

[0044] In a specific embodiment, the voice data to be translated obtained through a recording device may contain various noises, such as background noise, electronic equipment noise, wind noise, etc. These noises will seriously affect the accuracy of voice recognition and translation. Therefore, it is first necessary to perform noise reduction processing on the voice data to be translated, which can remove or reduce the impact of these noises, thereby improving the quality of the voice signal and making the subsequent voice recognition and translation processes more accurate.

[0045] Specifically, the speech feature data to be translated includes the amplitude of the speech signal, the amplitude of each frequency component of the speech signal, and the average frequency of the formant of the speech signal.

[0046] It needs to be explained that the amplitude of the above-mentioned speech signal can be obtained through audio processing software. In the audio processing software, the amplitude is usually displayed directly through a waveform graph, and the vertical axis of the waveform graph represents the amplitude. The amplitude of each frequency component of the speech signal is obtained by dividing the speech signal into multiple short-time frames and performing Fourier transform on each frame to obtain a spectrum graph of the frame. Each point in the spectrum graph represents the amplitude of the corresponding frequency component. The resonance peak of the speech signal is obtained by performing logarithmic transform and inverse Fourier transform on the spectrum of the speech signal to obtain the inverse spectrum of the signal. In the inverse spectrum, the resonance peak will appear as an obvious peak. By detecting these peaks, the frequency of the resonance peak can be extracted, thereby obtaining the average frequency of the resonance peak of the speech signal.

[0047] Furthermore, the first translation result data specifically includes a first speech translation rate and a speech translation duration.

[0048] It should be explained that the above-mentioned voice translation duration is achieved by setting a translation timer inside the voice translation device. From the time the user starts inputting voice to the time the system outputs the translated voice, the translation timer will continue to run throughout the whole process. Finally, the voice translation device will use the time displayed by the timer as the voice translation duration, thereby obtaining the voice translation duration.

[0049] The specific analysis process of the first speech translation rate is as follows:

[0050] Get the voice data to be translated.

[0051] It should be explained that the above-mentioned voice data to be translated is obtained through a voice translation device, which integrates technologies such as voice recognition, machine translation and voice synthesis. When a user inputs a voice to be translated, the voice recognition technology is first used to convert the voice into text, and then the machine translation technology is used to translate the text from the source language into the target language. Finally, the voice synthesis technology is used to convert the translated text into voice output in the target language. In this way, the voice input by the user into the voice translation device is the voice data to be translated.

[0052] The speech data to be translated is divided into speech words to obtain words in the speech to be translated, and the words in the speech to be translated are counted and recorded as the number of words to be translated, and the pronunciation of the words in the speech to be translated is matched with the preset pronunciations of each translatable word in the speech translation database, thereby counting the number of words that successfully match the pronunciation of the words in the speech to be translated with the preset pronunciations of each translatable word in the speech translation database, and recorded as the number of translatable words.

[0053] It should be explained that the specific division method of the above-mentioned speech data to be translated for speech word division is: extract features from the signal of the speech to be translated, these features usually include spectral features (such as Mel-frequency cepstral coefficients MFCC), energy features, fundamental frequency features, etc., these features can reflect the acoustic characteristics of speech, and use the extracted features to divide the continuous speech signal into multiple speech segments, each speech segment corresponds to a word or phrase. This step involves voice activity detection (VAD) technology, which is used to distinguish between speech and silence. For each speech segment, a speech recognition model (such as a hidden Markov model HMM, a deep neural network DNN, a recurrent neural network RNN ​​or a long short-term memory network LSTM, etc.) is used for word recognition. Finally, the recognized words are post-processed, including removing redundancy, correcting errors, etc., to improve word segmentation. The accuracy of the word is improved; the pronunciation of the word in the to-be-translated speech is matched with the preset pronunciations of each translatable word in the speech translation database. The specific matching process is: the pronunciation of the word in the to-be-translated speech is feature-aligned with the preset pronunciation of the word in the speech translation database, which usually involves a dynamic time warping (DTW) algorithm or a hidden Markov model (HMM), etc., which is used to align the features of the two speech signals in time. Then, the cosine similarity method is used to obtain the similarity between the aligned features, which is recorded as word similarity. The word similarity is compared with the set similarity threshold to determine whether the pronunciation of the word in the to-be-translated speech matches the pronunciation of a preset word in the speech translation database. If the word similarity exceeds the set similarity threshold, it is considered that the match is successful. Finally, the number of words that successfully match is counted and recorded as the number of translatable words.

[0054] The number of translatable words is compared with the number of words to be translated to obtain a first speech translation rate.

[0055] Furthermore, the speech translation quality index has the following specific analysis process:

[0056] The amplitude of each frequency component of the speech signal is squared and the obtained value is recorded as the power of each frequency component. The values ​​are added together to obtain the total power of the frequency components. The total number of frequency components is counted, and the total power of the frequency components is divided by the total number of frequency components to finally obtain the average power of the speech signal.

[0057] The first speech translation rate, speech translation duration, amplitude of the speech signal, average power and average frequency of the formant are comprehensively analyzed to obtain the speech translation quality index. The specific analysis formula is:

[0058]

[0059] In the formula, is the speech translation quality indicator, and e is a natural constant.

[0060] The first speech translation rate refers to the ratio of the number of translatable words to the number of words to be translated.

[0061] The speech translation accuracy rate preset for the speech translation database refers to a minimum value of the speech translation accuracy rate preset in the speech translation database.

[0062] It is the average power of the speech signal, which refers to the average value of the speech signal energy.

[0063] The reference power preset for the speech translation database refers to the energy value used for comparison and reference in speech signal processing.

[0064] It is the average frequency of the formant of the speech signal, which refers to the average value of multiple formant frequencies generated by the resonance characteristics where the energy is relatively concentrated in the sound spectrum.

[0065] The formant reference frequency preset for the speech translation database refers to a pre-set formant frequency value based on the analysis of a standard speech signal or specific language characteristics, which is used as a benchmark for measuring and evaluating the quality of an input speech signal.

[0066] The duration of voice translation refers to the time required to complete the process from input to output of translation results for a piece of voice data.

[0067] The defined duration of speech translation preset for the speech translation database refers to a pre-set time standard used to evaluate and ensure the response speed and efficiency of speech translation.

[0068] In this embodiment, when the average power of the speech signal deviates greatly from the reference power preset in the speech translation database, this usually means that the energy distribution of the input speech signal is significantly different from the expected or standard situation. The average power with a large deviation may cause part of the speech signal to be incorrectly recognized or ignored during the translation process. For example, if the average power is much higher than the reference power, the system may mistake the background noise for valid speech, resulting in an increase in the speech translation time and a direct decrease in the first speech translation rate. On the contrary, if the average power is much lower than the reference power, the system may not be able to capture enough speech information, which will also result in a decrease in the first speech translation rate. When the resonance peak deviates greatly from the reference resonance peak of the preset speech signal, the translation system may misjudge the pronunciation, increase the speech translation time, and mistakenly recognize some phonemes as other phonemes. This misjudgment will directly affect the first speech translation rate. Therefore, by analyzing the data of the speech translation quality indicator, the quality of the speech signal can be judged more accurately, thereby reducing translation errors caused by poor signal quality.

[0069] In this embodiment, by performing nonlinear analysis on the first speech translation rate, speech translation time, amplitude, average power and resonance peak of the speech signal, the complex and nonlinear characteristics in the speech signal can be captured more accurately. These characteristics often have an important impact on the understanding of sound quality, pitch and semantics in speech translation. Through nonlinear analysis, the changes in these characteristics can be more carefully depicted, thereby improving the analysis accuracy. Nonlinear analysis can better adapt to these dynamic changes and ensure accurate translation results by adjusting analysis parameters and translation strategies in real time.

[0070] In this embodiment, there is a close relationship between the above-mentioned first speech translation rate, speech translation time, amplitude, average power and formant data of the speech signal. If the amplitude of the speech signal is large and the average power is high, it means that the speech signal is clearer, which may make the translation system easier to recognize and process, thus increasing the first speech translation rate and reducing the speech translation time. When the average frequency of the formant is high, the energy of the frequency band is also relatively high, thereby contributing to the average power. In speech data processing, the average power of the speech signal and the average frequency data of the formant are usually interrelated, and they jointly describe the intensity change, energy distribution and sound quality characteristics of the speech signal. Through the comprehensive analysis of these parameters, the quality and characteristics of the speech signal can be more comprehensively understood, thereby providing strong support for applications such as speech translation and speech recognition.

[0071] In this embodiment, the speech translation quality index is compared with the speech translation quality index limit value preset in the speech translation database. If the speech translation quality index is greater than or equal to the speech translation quality index limit value preset in the speech translation database, it means that the quality of the first translation has reached the preset standard, that is, the translation result is accurate and fluent, with no obvious ambiguity or very little ambiguity, so no further ambiguity adjustment operation is required, and the first translation result can be output as the final translation result. If the speech translation quality index is less than the speech translation quality index limit value preset in the speech translation database, the speech feature data to be translated is preprocessed.

[0072] Specifically, the preprocessing operation is performed on the speech feature data to be translated, specifically, each speech feature data point to be translated in the speech feature data to be translated is cleaned, and finally the cleaned speech feature data to be translated is recorded as the adjusted speech data.

[0073] It needs to be explained that the above-mentioned speech feature data points to be translated are multiple data in the speech feature data to be translated, which are recorded as speech feature data points to be translated in the present embodiment. In the present implementation example, the speech feature data to be translated is preprocessed before the ambiguous translation adjustment operation is performed on the speech feature data to be translated, which can eliminate or reduce translation errors caused by data quality problems, making the subsequent ambiguous translation adjustment operation less susceptible to abnormal factors.

[0074] In this embodiment, the cleaning operation of the above-mentioned speech feature data points to be translated can use the speech recognition technology in the data processing device to perform preliminary recognition on the speech feature data points to be translated and automatically correct obvious recognition errors.

[0075] Furthermore, the adjusting of the characteristic data of the speech specifically includes adjusting the average power, the average frequency of the formant and the average bandwidth of the speech signal, and adjusting the clarity data of the speech signal.

[0076] It should be explained that the above adjustment of the average power of the speech signal is achieved by dividing the speech signal into frames and calculating the energy of each frame signal, that is, the integral or sum of the square of the signal. Finally, the energy of all frames is averaged to obtain the average power of the entire speech signal. The average frequency of the resonance peak of the speech signal can be adjusted by using spectrum analysis technology (such as fast Fourier transform FFT) to convert the speech signal from the time domain to the frequency domain. In the frequency domain, the energy distribution of the signal at different frequencies can be observed, so as to identify the position of the resonance peak. On the spectrum diagram, the resonance peak appears as a wide spectrum maximum or local maximum with concentrated energy. Specific algorithms (such as linear predictive coding LPC, cepstrum method, etc.) can be used to detect the position and frequency of these resonance peaks. The average bandwidth of the speech signal can be adjusted by using audio processing software to obtain the average bandwidth of the speech signal.

[0077] The adjusting the clarity data of the speech signal specifically includes adjusting the amplitude and fundamental frequency of the speech signal and adjusting the speech speed of the speech.

[0078] It should be explained that the amplitude of the voice signal can be adjusted through audio processing software. In the audio processing software, the amplitude is usually displayed directly through a waveform graph, and the vertical axis of the waveform graph represents the amplitude. The fundamental frequency of the voice signal can be adjusted through audio editing software. Usually, audio editing software provides pitch analysis or fundamental frequency extraction functions, which can directly obtain the fundamental frequency value. The speech speed is adjusted by calculating the duration of the voice signal and the amount of voice content.

[0079] Specifically, the speech translation adjustment index has a specific analysis process as follows:

[0080] The amplitude, fundamental frequency and speaking speed of the adjusted speech signal are comprehensively analyzed to obtain the clarity index of the adjusted speech. The specific analysis formula is:

[0081]

[0082] In the formula, To adjust the clarity index of speech.

[0083] To adjust the amplitude of the speech signal refers to adjusting the volume of the speech signal.

[0084] The influence factor corresponding to the amplitude unit value preset in the speech translation database can be directly obtained from the speech translation database. The corresponding relationship can be a preset mapping relationship. For example, a mapping set is formed according to the amplitude of the historically adjusted speech signal and the influence factor corresponding to the amplitude unit value of the speech signal preset in the speech translation database, and the amplitude mapping set of the real-time adjusted speech signal is used to obtain the influence factor corresponding to the amplitude unit value. The mapping relationship can be one-to-one or many-to-one. At the same time, in this example, its value range is [0, 1].

[0085] The influence factor corresponding to the fundamental frequency unit value preset in the speech translation database can be directly obtained from the speech translation database. The corresponding relationship can be a preset mapping relationship. For example, a mapping set is formed according to the fundamental frequency of the historically adjusted speech signal and the influence factor corresponding to the fundamental frequency unit value of the adjusted speech signal preset in the speech translation database, and the fundamental frequency mapping set of the real-time adjusted speech signal is used to obtain the influence factor corresponding to the fundamental frequency unit value. The mapping relationship can be one-to-one or many-to-one. At the same time, in this example, its value range is [0, 1].

[0086] To adjust the fundamental frequency of the speech signal, it refers to the lowest frequency of periodic oscillation in the speech waveform.

[0087] The correction factor corresponding to the speech rate preset in the speech translation database can be directly selected from the correction factor corresponding to the speech rate preset in the speech translation database, and the corresponding relationship can be a preset mapping relationship. For example, a mapping set is formed based on the speech rate of the historically adjusted speech signal and the correction factor corresponding to the speech rate, and the speech rate mapping set of the real-time adjusted speech signal is used to obtain the correction factor corresponding to the speech rate. The mapping relationship can be one-to-one or many-to-one. At the same time, in this example, its value range is [0, 1].

[0088] To adjust the speaking speed of speech refers to adjusting the rhythm of speaking.

[0089] To adjust the reference speaking speed of speech refers to adjusting the reference rhythm of speech.

[0090] In this embodiment, a lower amplitude means a lower loudness of the sound, which may make the speech signal difficult to hear and recognize in the current situation, thereby reducing the clarity index. The fundamental frequency corresponds to the frequency of the vocal cord vibration and represents the pitch of the sound. The faster the vocal cord vibrates, the higher the fundamental frequency. A lower fundamental frequency will cause the speech to sound dull and reduce the efficiency of speech reception. When the speed of the adjusted speech deviates greatly from the reference speed of the adjusted speech, it means that the current speaking speed is too fast. A fast speaking speed may cause key information and phonemes in the speech to be ignored or blurred, making it difficult to capture important information, which in turn leads to ambiguity in speech translation. Therefore, by analyzing the various parameters in the clarity index of the adjusted speech, it is helpful to maintain the naturalness and coherence of the language, so that the translated speech is more in line with the expression habits of the target language, and further improve the translation efficiency.

[0091] The average power, average frequency of the formant and average bandwidth of the adjusted speech signal are comprehensively analyzed to obtain the speech translation adjustment index. The specific analysis formula is:

[0092]

[0093] In the formula, is the index for adjusting the speech translation, e is a natural constant, To adjust the speech clarity index, To adjust the average power of the speech signal refers to changing the average value of the energy consumption of the speech signal per unit time.

[0094] The reference power of the adjusted speech signal preset for the speech translation database refers to the reference standard for the average power of the speech signal preset in the database in order to optimize the translation effect and maintain the naturalness and intelligibility of the output speech in the speech translation system.

[0095] To adjust the average frequency of the formant of the speech signal refers to the average value of multiple formant frequencies generated by the resonance characteristics in which the spectrum energy in the speech signal is relatively concentrated.

[0096] The reference frequency of the adjusted speech signal formant preset for the speech translation database refers to a reference standard preset in the database regarding the average frequency characteristic of the speech signal formant.

[0097] The correction factor corresponding to the average power of the adjusted speech signal is preset in the speech translation database. The correction factor corresponding to the average power of the adjusted speech signal can be directly obtained from the speech translation database, and the corresponding relationship can be a preset mapping relationship. For example, a mapping set is formed based on the average power of the historically adjusted speech signal and the correction factor corresponding to the average power of the adjusted speech signal preset in the speech translation database, and the real-time average power mapping set of the adjusted speech signal is used to obtain the correction factor corresponding to the average power of the adjusted speech signal. The mapping relationship can be one-to-one or many-to-one. At the same time, in this example, its value range is [0, 1].

[0098] The correction factor corresponding to the average formant frequency of the adjusted speech signal preset in the speech translation database can be directly obtained from the speech translation database. The corresponding relationship can be a preset mapping relationship. For example, a mapping set is formed based on the historical average formant frequency of the adjusted speech signal and the correction factor corresponding to the average formant frequency of the adjusted speech signal preset in the speech translation database, and the correction factor corresponding to the average formant frequency of the adjusted speech signal is obtained from the real-time mapping set of the average formant frequency of the adjusted speech signal. The mapping relationship can be one-to-one or many-to-one. At the same time, in this example, its value range is [0, 1].

[0099] To adjust the average bandwidth of a speech signal refers to changing the average width occupied by the speech signal in the frequency domain.

[0100] The influence factor corresponding to the average bandwidth unit value of the adjusted voice signal preset in the voice translation database can be directly obtained from the voice translation database. The corresponding relationship can be a preset mapping relationship. For example, a mapping set is formed based on the average bandwidth of the historical adjusted voice signal and the influence factor corresponding to the average bandwidth unit value of the adjusted voice signal preset in the voice translation database, and the real-time average bandwidth mapping set of the adjusted voice signal is used to obtain the influence factor corresponding to the average bandwidth unit value of the adjusted voice signal. The mapping relationship can be one-to-one or many-to-one. At the same time, in this example, its value range is [0, 1].

[0101] The weight factor corresponding to the clarity index of the adjusted speech preset in the speech translation database can be directly obtained from the speech translation database. The corresponding relationship can be a preset mapping relationship. For example, a mapping set is formed based on the clarity index of the historical adjusted speech signal and the weight factor corresponding to the clarity index of the adjusted speech preset in the speech translation database, and the real-time clarity index mapping set of the adjusted speech signal is used to obtain the weight factor corresponding to the clarity index of the adjusted speech. The mapping relationship can be one-to-one or many-to-one. At the same time, in this example, its value range is [0, 1].

[0102] In this embodiment, if the deviation between the average power of the adjusted speech signal and the preset reference power of the adjusted speech signal is large, the intelligibility of the speech translation may be reduced. The purpose of speech translation is to ensure that the translated speech can be accurately understood by the listener, resulting in changes in the spectral characteristics of the speech signal, thereby affecting the clarity and recognition of the speech. When the deviation between the average frequency of the resonance peaks of the adjusted speech signal and the reference resonance peaks of the adjusted speech signal preset in the speech translation database is large, the sound quality of the translated speech signal will change significantly, and the spectral characteristics of the speech signal will be changed, making it difficult for the recognition algorithm to accurately capture the effective features in the speech, which will lead to a decrease in the speech recognition rate. Ambiguity still occurs in the adjustment result, affecting the accuracy of the speech translation. When the average bandwidth of the adjusted speech signal is small, some frequency components of the speech signal may be lost, which will reduce the information available to the speech recognition algorithm, thereby reducing the accuracy of recognition. When the clarity index of the adjusted speech is low, the speech translation device cannot accurately recognize words and phrases, increasing the risk of recognition errors. Therefore, by analyzing the various data in the speech translation adjustment index, the ambiguity in the speech translation can be effectively identified and reduced, thereby improving the accuracy of the translation.

[0103] In this embodiment, by performing nonlinear analysis on the average power, average frequency of the resonance peaks, and average bandwidth of the adjusted speech signal, it is possible to fit curves or non-linear relationships that cannot be captured by linear regression. This is applicable to various complex relationships in speech signals, such as the nonlinear correlation between frequency and energy. At the same time, since nonlinear data analysis can adapt to the complexity of the data, it can usually provide more accurate speech translation adjustment indicators.

[0104] In this embodiment, the reference power value of the adjusted speech signal is set to 50 decibels, the reference resonance peak average frequency value of the adjusted speech signal is set to 500 Hz, the correction factor corresponding to the average power of the adjusted speech signal is 0.3, the correction factor corresponding to the resonance peak average frequency of the adjusted speech signal is 0.4, the influence factor corresponding to the average bandwidth unit value of the adjusted speech signal is 0.3, and the weight factor corresponding to the clarity index is 0.4. Thus, a data relationship table between the speech translation adjustment index and each parameter can be obtained, as shown in Table 1 Speech Translation Adjustment Index Table.

[0105] Table 1 Speech translation adjustment index table

[0106]

[0107] By observing the data in the speech translation adjustment index table in Table 1 above, it is found that the speech translation adjustment index will change according to the changes in the average power, average resonance peak frequency, average bandwidth of the adjusted speech signal and the clarity index of the adjusted speech. The larger the deviation value of the average power and average resonance peak frequency of the adjusted speech signal from the reference power and reference resonance peak average frequency of the adjusted speech signal, the larger the speech translation adjustment index obtained. At the same time, the larger the value of the average bandwidth of the adjusted speech signal and the larger the clarity index of the adjusted speech, the larger the speech translation adjustment index obtained.

[0108] In this embodiment, there is a close relationship between the above-mentioned average bandwidth of the adjusted speech signal and the average frequency of the formant of the adjusted speech signal. The relationship between the average bandwidth of the adjusted speech signal and the average frequency of the formant is an important aspect of speech signal processing. The average bandwidth and the average frequency of the formant are both key parameters for describing the characteristics of the speech signal, and there is a mutually influencing relationship between them. Therefore, a simulation curve between the average bandwidth of the adjusted speech signal and the average frequency of the formant of the adjusted speech signal can be constructed, as shown in the following example. Figure 3 The simulation curve between the average bandwidth and the average frequency of the resonance peak is shown in the figure, where the horizontal axis is the average bandwidth in Hertz, and the vertical axis is the average frequency of the resonance peak in Hertz.

[0109] By observation Figure 3The simulation curve diagram between the average bandwidth and the average frequency of the resonance peak shows that when the average bandwidth of the speech signal changes, it will directly affect the distribution of frequency components in the signal, and then change the frequency and amplitude of the average frequency of the resonance peak. If the average bandwidth becomes narrower, that is, the high-frequency components are removed, then the average frequency of the high-frequency resonance peak may decrease because the high-frequency components are filtered out. On the contrary, if the average bandwidth becomes wider, that is, more high-frequency components are retained, then the average frequency of the high-frequency resonance peak may increase, making the sound sound brighter and clearer.

[0110] In this embodiment, the speech translation adjustment index is compared with the speech translation adjustment index threshold preset in the speech translation database. If the speech translation adjustment index is greater than or equal to the preset speech translation adjustment index threshold, it means that the quality of the speech signal after the ambiguity adjustment is high, and the uncertainty or ambiguity therein has been effectively reduced or eliminated. Therefore, this segment of speech signal will not be greatly affected during translation, and the translation rate will be less affected. Then, the adjusted speech data after the ambiguous translation adjustment is completed is translated for the second time. On the contrary, if the speech translation adjustment index is less than the preset speech translation adjustment index threshold, it means that after adjusting the ambiguous speech data, there are still ambiguity problems that cannot be solved by speech translation. These problems are usually caused by the poor quality of the translated speech input by the user (such as too much noise, too fast speech speed, unclear pronunciation, etc.). These factors will make the speech recognition and translation process difficult. Then, the current speech is directly fed back through the speech translation feedback device. The specific feedback can be to point out specific problems in the translation, such as improper word selection, grammatical errors, adjusting speech speed, clarity, or selecting a quieter environment for re-input.

[0111] Specifically, the feedback is provided to the speech ambiguous translation adjustment process, and the specific feedback process is:

[0112] After the ambiguous translation adjustment is completed, the adjusted speech data is translated for the second time to obtain second translation result data, which is then comprehensively analyzed with the speech translation adjustment index to obtain the ambiguous translation adjustment accuracy index.

[0113] The second translation result data specifically includes a second voice-adjusted translation rate.

[0114] The second voice adjusts the translation rate, and the specific analysis process is as follows:

[0115] Get the adjusted voice data.

[0116] It should be explained that the above-mentioned adjusted speech completion data refers to the adjusted speech data after the ambiguous translation adjustment is completed.

[0117] The adjusted speech completion data is divided into speech words to obtain words in the adjusted speech completion data, and the words in the adjusted speech completion data are counted and recorded as the number of adjusted words to be translated, and the pronunciation of the words in the adjusted speech completion data is matched with the preset pronunciations of each translatable word in the speech translation database, thereby counting the number of words that successfully match the pronunciation of the words in the adjusted speech completion data with the preset pronunciations of each translatable word in the speech translation database, which is recorded as the number of adjusted translatable words.

[0118] It should be explained that the above-mentioned method of adjusting the speech completion data for speech word division is the same as the method of dividing the speech data to be translated, and the process of adjusting the word pronunciation in the speech completion data to match the preset pronunciations of each translatable word in the speech translation database is the same as the process of matching the word pronunciation in the above-mentioned speech to be translated with the preset pronunciations of each translatable word in the speech translation database.

[0119] The adjusted number of translatable words and the adjusted number of words to be translated are ratio-processed to obtain the adjusted translation rate of the second voice.

[0120] The ambiguous translation adjustment accuracy index, the specific analysis process is as follows:

[0121] The second voice adjustment translation rate and voice translation adjustment index are comprehensively analyzed to obtain the ambiguous translation adjustment accuracy index. The specific analysis formula is:

[0122]

[0123] In the formula, Adjust accuracy metrics for ambiguous translations, Adjust metrics for speech translation.

[0124] Adjusting the translation rate for the second voice refers to adjusting the ratio of the number of translatable words to the number of words to be translated.

[0125] The voice-adjusted allowable translation rate preset for the voice translation database refers to the voice-adjusted minimum translation rate preset for the voice translation database.

[0126] The weight factors corresponding to the speech translation adjustment indicators preset in the speech translation database can be directly obtained from the speech translation database. The corresponding relationship can be a preset mapping relationship. For example, a mapping set is formed based on the weight factors corresponding to the historical speech translation adjustment indicators and the speech translation adjustment indicators, and the real-time speech translation adjustment indicator mapping set is used to obtain the weight factors corresponding to the speech translation adjustment indicators. The mapping relationship can be one-to-one or many-to-one. At the same time, in this example, its value range is [0, 1].

[0127] In this embodiment, a lower speech translation adjustment index means that the ambiguity in the speech is not properly handled, which will directly cause the translation result to deviate from the original semantics, resulting in a lower second speech adjustment translation rate. When it is even lower than the preset speech adjustment allowable translation rate, it means that the ambiguity in the speech is not adjusted properly, which will cause the speech translation device to be unable to accurately understand the intention and meaning of the original speech, directly causing the translation result to deviate from the original semantics and reduce the accuracy of the translation. Therefore, through a detailed analysis of each parameter in the ambiguous translation adjustment accuracy index, the quality of speech translation can be improved and the translation efficiency can be improved.

[0128] The ambiguous translation adjustment accuracy index is compared with the ambiguous translation adjustment accuracy index threshold value preset in the speech translation database. If the ambiguous translation adjustment accuracy index of the second translation is greater than or equal to the ambiguous translation adjustment accuracy index threshold value preset in the speech translation database, there is no need to provide feedback on the speech ambiguous translation adjustment process through the speech translation feedback device.

[0129] It needs to be explained that when the ambiguous translation adjustment accuracy index is greater than or equal to the ambiguous translation adjustment accuracy index threshold, it means that the processing result of the speech signal in the process of adjusting the current ambiguous speech data has reached an acceptable and relatively stable state, and the translation quality in this state accurately meets or exceeds the preset standard. Therefore, there is no need to provide feedback on the speech ambiguity through the speech translation feedback device, and the second translation result is output as the final translation result.

[0130] If the ambiguous translation adjustment accuracy index is less than the ambiguous translation adjustment accuracy index threshold preset in the speech translation database, the speech ambiguous translation adjustment process is fed back through the speech translation feedback device.

[0131] It needs to be explained that when the speech translation adjustment index is less than the preset threshold of the ambiguous translation adjustment accuracy index, it means that there are still ambiguous problems that cannot be solved by speech translation through the adjustment of the ambiguous speech data. Due to the complexity of the speech itself, there will still be ambiguities that cannot be completely eliminated, resulting in the speech translation output results being not only inaccurate, but also semantically ambiguous or having multiple interpretations. In this case, it is particularly important to provide feedback on speech ambiguity through a speech translation feedback device. The specific feedback can be suggestions to update the translation vocabulary or optimize the language model.

[0132] Reference Figure 2 As shown, the second aspect of the present invention provides a device for a speech translation method for resolving speech ambiguity, comprising: a recording device, a data processing device, a speech translation device, and a speech translation feedback device;

[0133] The recording device is used to obtain voice data that needs to be translated.

[0134] The data processing device is used to receive voice data to be translated through the Internet of Things, and preprocess to obtain voice data to be translated, receive and preprocess translated voice feature data in the initial translated voice data of the voice translation device, obtain adjusted voice data, respectively analyze the first translation result data and the voice feature data to be translated, and finally obtain a voice translation quality index, and analyze the feature data of the adjusted voice to obtain a voice translation adjustment index.

[0135] The speech translation device is used to receive speech data to be translated from a data processing device through the Internet of Things and translate it to obtain first translation result data, and receive and adjust the adjusted speech data of the data processing device.

[0136] The speech translation feedback device is used to receive the speech translation quality index and the speech translation adjustment index of the data processor through the Internet of Things, and finally provide feedback on the speech ambiguity.

[0137] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined in this specification, they should all fall within the protection scope of the present invention.

Claims

1. A speech translation method for resolving speech ambiguity, characterized in that: include: S1. Acquiring voice data: The data processing device acquires the voice data to be translated through the recording device, and preprocesses the voice data to be translated, and the preprocessed voice data to be translated is recorded as voice data to be translated; S2. Speech data analysis: The speech translation device performs a first speech translation on the speech data to be translated, and the data processing device extracts the translation result corresponding to the first speech translation to obtain the first translation result data and the speech feature data to be translated, and performs a comprehensive analysis to obtain the speech translation quality index; S3. Translation speech ambiguity adjustment: compare the speech translation quality index with the speech translation quality index boundary value preset in the speech translation database. If the speech translation quality index is less than the speech translation quality index boundary value preset in the speech translation database, perform a preprocessing operation on the speech feature data to be translated, and record the preprocessed speech feature data to be translated as the adjusted speech data. Thus, the speech translation device performs an ambiguous translation adjustment operation on the adjusted speech data, and records the speech data after the ambiguous translation adjustment operation as the adjusted speech data. Based on the ambiguous translation adjustment process, extract the feature data of the adjusted speech, obtain the speech translation adjustment index through analysis, and compare it with the preset speech translation adjustment index threshold. Based on the comparison result, perform a second translation on the adjusted speech data after the ambiguous translation adjustment is completed to obtain the second translation result data, and perform a comprehensive analysis with the speech translation adjustment index to obtain the ambiguous translation adjustment accuracy index. Compare the ambiguous translation adjustment accuracy index with the ambiguous translation adjustment accuracy index threshold preset in the speech translation database, and finally provide feedback on the speech ambiguous translation adjustment process through the speech translation feedback device. The speech feature data to be translated specifically includes the amplitude of the speech signal, the amplitude of each frequency component of the speech signal, and the average frequency of the formant of the speech signal; The specific analysis process of the speech translation quality indicator is as follows: The amplitude of each frequency component of the speech signal is squared and the obtained value is recorded as the power of each frequency component, and then added to obtain the total power of the frequency component. The total number of frequency components is counted, and then the total power of the frequency component is divided by the total number of frequency components to finally obtain the average power of the speech signal; Comprehensively analyzing the first speech translation rate, speech translation duration, amplitude, average power, and average frequency of formant of the speech signal to obtain a speech translation quality index; The specific analysis process of the speech translation adjustment index is as follows: Comprehensively analyzing the amplitude and fundamental frequency of the adjusted speech signal and the speaking speed of the adjusted speech to obtain the clarity index of the adjusted speech; The average power, average frequency of formant, average bandwidth and clarity index of the adjusted speech signal are comprehensively analyzed to obtain the speech translation adjustment index.

2. A speech translation method for resolving speech ambiguity according to claim 1, characterized in that: The preprocessing operation on the speech data to be translated is specifically to monitor the noise of each speech data point in the speech data to be translated to obtain the noise value of each speech data point. If the noise value of a speech data point is higher than the noise reference threshold preset in the speech translation database, the speech data point is subjected to noise reduction processing. Thus, the statistics of each speech data point on which noise reduction is completed and each speech data point on which noise reduction is not performed are counted, which is the speech data to be translated after the final preprocessing is completed and recorded as the speech data to be translated.

3. A speech translation method for resolving speech ambiguity according to claim 1, characterized in that: The first translation result data specifically includes a first speech translation rate and a speech translation duration; The specific analysis process of the first speech translation rate is as follows: Obtain the voice data to be translated; The speech data to be translated is divided into speech words to obtain words in the speech to be translated, and the words in the speech to be translated are counted as the number of words to be translated, and the pronunciation of the words in the speech to be translated is matched with the preset pronunciations of each translatable word in the speech translation database, thereby counting the number of words that successfully match the pronunciation of the words in the speech to be translated with the preset pronunciations of each translatable word in the speech translation database, and recording it as the number of translatable words; The number of translatable words is compared with the number of words to be translated to obtain a first speech translation rate.

4. A speech translation method for resolving speech ambiguity according to claim 1, characterized in that: The preprocessing operation of the speech feature data to be translated specifically involves cleaning each speech feature data point in the speech feature data to be translated, and finally recording the cleaned speech feature data to be translated as adjusted speech data.

5. A speech translation method for resolving speech ambiguity according to claim 1, characterized in that: The adjusting of the characteristic data of the speech specifically includes adjusting the average power, the average frequency of the formant and the average bandwidth of the speech signal, and adjusting the clarity data of the speech signal; The adjusting the clarity data of the speech signal specifically includes adjusting the amplitude and fundamental frequency of the speech signal and adjusting the speech speed of the speech.

6. A speech translation method for resolving speech ambiguity according to claim 1, characterized in that: The specific analysis process of the speech translation adjustment index is as follows: In the formula, is the index for adjusting the speech translation, e is a natural constant, To adjust the average power of the speech signal, The preset reference power of the adjusted speech signal for the speech translation database, The correction factor corresponding to the average power of the adjusted speech signal preset in the speech translation database, To adjust the average frequency of the formant of the speech signal, Adjust the speech signal formant reference frequency preset for the speech translation database, The correction factor corresponding to the average frequency of the formant of the speech signal preset in the speech translation database, To adjust the average bandwidth of the speech signal, The impact factor corresponding to the bandwidth unit value of the adjusted speech signal preset in the speech translation database, To adjust the speech clarity index, The weight factor corresponding to the speech clarity index preset in the speech translation database is adjusted.

7. A speech translation method for resolving speech ambiguity according to claim 1, characterized in that: The feedback on the speech ambiguous translation adjustment process is specifically as follows: The adjusted speech data after the ambiguous translation adjustment is completed is translated for the second time to obtain second translation result data, and is comprehensively analyzed with the speech translation adjustment index to obtain the ambiguous translation adjustment accuracy index; The ambiguous translation adjustment accuracy index is compared with the ambiguous translation adjustment accuracy index threshold value preset in the speech translation database, and if the ambiguous translation adjustment accuracy index is greater than or equal to the ambiguous translation adjustment accuracy index threshold value preset in the speech translation database, there is no need to provide feedback on the speech ambiguous translation adjustment process through the speech translation feedback device; If the ambiguous translation adjustment accuracy index is less than the ambiguous translation adjustment accuracy index threshold preset in the speech translation database, the speech ambiguous translation adjustment process is fed back through the speech translation feedback device.

8. A device for applying the speech translation method for resolving speech ambiguity as claimed in any one of claims 1 to 7, characterized in that: include: Recording device, data processing device, speech translation device and speech translation feedback device; The recording device is used to obtain voice data that needs to be translated; The data processing device is used to receive voice data to be translated through the Internet of Things, and perform preprocessing to obtain voice data to be translated, receive and preprocess translated voice feature data in the initial translated voice data of the voice translation device, obtain adjusted voice data, analyze the first translation result data and the voice feature data to be translated, and finally obtain a voice translation quality index, and analyze the feature data of the adjusted voice to obtain a voice translation adjustment index; The speech translation device is used to receive the speech data to be translated from the data processing device through the Internet of Things and translate it to obtain first translation result data, and receive and adjust the adjusted speech data from the data processing device; The speech translation feedback device is used to receive the speech translation quality index and the speech translation adjustment index of the data processing device through the Internet of Things, and finally provide feedback on the speech ambiguous translation adjustment process.

Images