82results about How to "Reduce false wake-up rate" patented technology

The invention discloses a wake-up word preset confidence threshold adjustment method and system. The method comprises steps that S11, daily use behavior data of a user is obtained; S12, the wake-up word preset confidence threshold is adjusted according to the daily use behavior data. The method is advantaged in that the wake-up word preset confidence threshold is dynamically set and adjusted through acquiring the daily usage behavior data of the user, personal use characteristics of the user are fully considered, firstly, whether the preset confidence threshold is lowered overall is determinedthrough the daily behavior confidence average value of the user, the preset confidence threshold is further adjusted based on the use number for each time period, the device wake-up rate is effectively improved, and the false wake-up rate is reduced.

The embodiment of the invention provides a voice wake-up method and a voice interaction method. The voice wake-up method comprises the steps of receiving a voice input signal; determining the similarity of the voice input signal and a preset wake-up voice signal exceeds a first preset threshold or not; if so, sending the voice input signal to a cloud server so as to enable the cloud server to perform voice recognition on the voice input signal and return a recognition result; receiving the recognition result which is returned by the cloud server and corresponds to the voice input signal; determining whether the similarity of the recognition result and the preset wake-up voice signal exceeds a second preset threshold or not, wherein the second preset threshold is greater than the first preset threshold; and if so, waking up a voice interaction function. The voice wake-up method and the voice interaction method provided by the embodiment of the invention can give consideration to the requirements of low power consumption and low error wake-up rate.

The invention discloses a voice wake-up method and device. The voice wake-up method comprises steps as follows: acquiring a to-be-processed voice signal; decoding the voice signal according to a pre-generated searching space to obtain a voice recognition result, wherein the searching space comprises a path where an inverse model is located, and the inverse model comprises a first inverse model body generated by training word segmentation results of wake-up words; judging whether the preset number of characters contains at least part of characters in the wake-up words when the preset number of first characters in the voice recognition result is obtained: directly determining cancellation of wake-up operation if at least part of characters in the wake-up words are not contained, and ending decoding of the voice signal. The method has the advantages that the mistaken wake-up rate can be reduced, power consumption can be reduced, and the like.

The embodiment of the invention discloses a voice wakeup method. The voice wakeup method comprises the following steps: S101, judging that whether the preset conditions are satisfied or not, if so, carrying out step 103, otherwise, carrying out step 105; S103, activating the operation of waking up voice control logic without a wake-up word; and S105, not activating the operation of waking up voicecontrol logic without the wake-up word. Through the voice wakeup method, the flow of voice interaction between a user and the voice control logic can be simplified, so that the voice interaction is more convenient and rapid, more humanized and more intelligent, the voice interaction efficiency is improved, and the experience degree of the user is improved.

The invention discloses a voice awakening optimization method based on cascade DNN. The method comprises the steps that 1, a voice signal is acquired by a microphone in real time, and feature extraction is conducted to obtain frame-by-frame acoustic features of the real-time voice signal; 2, an acoustic feature sequence is intercepted according to a fixed window length to form a frame serving as first-stage DNN input; 3, forward process calculation of a first-stage DNN acoustic model is carried out, and the acoustic posterior probability of frame-by-frame phonemes is obtained through output; 4, first-stage DNN output is intercepted according to the fixed window length to form a one-frame phoneme posterior probability sequence to serve as second-stage DNN input; 5, second-stage DNN forwardprocess calculation is carried out for judgment, and whether awakening is carried out or not is output. The anti-noise capability of the DNN can be utilized to the maximum extent, the environmental adaptability is high, and the situation that a VAD is manufactured first and then awakening detection is carried out is not needed; a voice background does not need to be independently modeled; the twostages of models can be complementary, so that corpora required by training can be greatly reduced; no language model is generated, and text corpora are not needed.

The invention relates to a voice wake-up method. The method comprises the steps: receiving a plurality of voice wake-up messages sent by a plurality of electronic devices, wherein each voice wake-up message comprises a distance and a wake-up energy value, the distance represents the distance between the electronic devices sending the voice wake-up messages and a sound source, and the wake-up energy value represents the energy value of a wake-up word sent by the sound source and received by the electronic devices sending the voice wake-up messages; according to the distances and the wake-up energy values in the plurality of voice wake-up messages of the plurality of electronic devices, determining whether the energy attenuation of the wake-up word emitted by the sound source conforms to theattenuation rule of the sound energy radiated by a point sound source; and when it is determined that the energy attenuation of the wake-up word emitted by the sound source does not accord with the attenuation rule of the sound energy radiated by the point sound source, sending a wake-up forbidding instruction to the plurality of electronic devices. According to the scheme provided by the invention, the false awakening rate of the device can be reduced.

The invention provides voice equipment and a control method, device and equipment thereof, and a storage medium. The method comprises the following steps of acquiring wake-up failure frequency of thevoice equipment in a preset time period, if the wake-up failure frequency is greater than or equal to a first preset frequency, reducing a wake-up threshold corresponding to wake-up information of thevoice equipment, if the wake-up failure frequency is smaller than or equal to a second preset frequency, increasing a wake-up threshold corresponding to wake-up information of the voice equipment, and if the wake-up failure frequency is less than the first preset frequency and greater than the second preset frequency, maintaining a wake-up threshold corresponding to wake-up information of the voice equipment, thereby realizing automatic adjustment of the wake-up threshold of the voice equipment in any usage scenario of the user and in the speaking volume of the user habit. By adopting the technical scheme of the invention, the wake-up rate of the voice equipment can be improved at a relatively low false wake-up rate.

A two-stage wakeup method and a two-stage wakeup device applied to an ambiguity path recognition system are disclosed. The method comprises the following steps: a timed wakeup control module sends a first cycle enable signal to a radio-frequency module; the radio-frequency module receives an external wakeup signal and sends a characterization signal of the external wakeup signal to the timed wakeup control module within the interception time of the first cycle enable signal; when determining that the characterization signal meets a first threshold condition, the timed wakeup control module stops sending the first cycle enable signal, and sends a second cycle enable signal to the radio-frequency module; the radio-frequency module sends the characterization signal of the received external wakeup signal to the timed wakeup control module within the interception time of the second cycle enable signal; and the timed wakeup control module outputs an internal wakeup signal to a system control module when determining that the characterization signal meets a second threshold condition. By adopting the method, the power consumption of the system is reduced greatly. In addition, the device of the invention can achieve all the beneficial effects of the method.

The invention provides a voice activity detection method, which comprises the following steps of obtaining voice signals collected by a microphone and vibration signals collected by a sensor; converting the voice signals from a time domain to a frequency domain; converting the vibration signals from the time domain to the frequency domain; calculating the frequency domain correlation coefficient by using the voice signal and the vibration signals; calculating the detection quantity in the detection frequency range; when the detection quantity is higher than a preset threshold value, judging the voice signals collected by the microphone to be speaking voice of a wearer; otherwise, judging the voice signals collected by the microphone to have noise or interference signals. By calculating thefrequency domain correlation coefficient, the detection quantity is calculated in the detection frequency range, the voice activity detection accuracy is improved; the noise inhibition definition andthe intelligibility of the noise inhibition are improved.

The invention relates to the field of artificial intelligence, and discloses a wake-up model training method. The method comprises the steps of extracting audio frames from specified voice statementsin a training set to obtain an acoustic feature matrix; inputting the acoustic feature matrix into a keyword detector of a first model to obtain a first spatial feature, inputting the acoustic featurematrix into an encoder of a second model to obtain a second spatial feature, the first model being a wake-up model to be trained, and the second model being a trained noise reduction model; calculating the difference between the spatial features of the first spatial feature and the second spatial feature; according to the calculation mode of the difference of the spatial features corresponding tothe specified voice statement, calculating the difference of the spatial features corresponding to all voice statements in the training set; and forming a loss function training wake-up model for training the wake-up model according to the differences of the spatial features corresponding to all the voice sentences and the preset cross entropy loss of the wake-up model. And the feature vector ofthe high-dimensional space is used as a knowledge distillation sample to assist in training a wake-up model, so that the wake-up effect is improved.

The invention provides a wake-up model updating method. The method comprises the following steps of: separating a command word and a wake-up word in original voice data; determining the probability ofcommand words according to the language model, and deleting the command words with the probability lower than a first threshold value in the command words; and inputting the wake-up word into a wake-up model, determining a score output by the wake-up model, and deleting the wake-up word with the score lower than a second threshold value in the wake-up word. Therefore, the built wake-up model withhigher false wake-up probability is updated, command words and wake-up words are screened respectively, and the wake-up words are closer to preset wake-up words by eliminating false wake-up data or distortion data. Meanwhile, the command words serve as training data, the number of the training data is guaranteed, and the false awakening probability of the updated awakening model is lower.

The invention provides a voice interruption wake-up method and device and a computer readable storage medium. The voice interruption wake-up method is applied to an MCU of a voice electronic device. The method comprises the following steps: obtaining a state parameter and an environment parameter of the current voice electronic device, and the environment parameter comprises external voice data; wherein the state parameters comprise a networking state; executing preset response setting based on the acquired state parameters and the environment parameters; and identifying and analyzing the external voice data, and if identifying and analyzing that the external voice data contains a preset wake-up word, controlling to wake up own voice interaction service. The voice interruption wake-up method is applied to the MCU of the voice electronic device, the interruption wake-up effect of an MCU small system is improved by obtaining the networking state and the voice data and adjusting accordingto the networking state, so that the high wake-up success rate is guaranteed, the false wake-up rate is reduced, and the user experience is improved.

The invention discloses a voice wake-up system based on environment detection. The system comprises an environment detection module, a wake-up module and a comprehensive judgment module, wherein the environment detection module and the wake-up module are electrically connected with the comprehensive judgment module respectively; the environment detection module is used for detecting the ambient environment; the wake-up module is used for recognizing a voice signal received by intelligent equipment; when a wake-up word is recognized, a wake-up request is sent to the comprehensive judgment module; the comprehensive judgment module is used for receiving the wake-up request sent by the wake-up module, acquiring data sent by the environment detection module and judging whether intelligent equipment is waken up or not. After the system detects a voice wake-up command, whether the intelligent equipment is waken up or not is determined through a method for detecting whether people exist in theambient environment and whether the time is daytime or not or a lamp is turned on or not; the incorrect wake-up rate of the voice wake-up system is reduced and the user experience is improved.

The invention discloses a speech wake-up method and device, electronic equipment and a computer readable storage medium. The method comprises the following steps: acquiring a speech signal; inputtingthe speech signal into a first model, and outputting a first posterior probability of each frame of phoneme in the speech signal through the first model; taking a speech signal corresponding to the phoneme of which the first posterior probability meets a first wake-up condition as a wake-up speech segment; inputting the spectrogram and the posterior probability graph of the wake-up voice segment into a second model, and outputting a second posterior probability of each frame of phoneme in the wake-up speech segment through the second model; if the second posterior probability satisfies a second wake-up condition, performing a wake-up operation. According to the embodiment of the invention, through secondary wake-up confirmation, the false wake-up rate can be reduced, and the wake-up accuracy is improved.

The invention provides a voice wake-up model training method and a voice false wake-up detection method and device, in a model training stage, a basic network and a classification layer are recognition network structures of an existing voice wake-up model, and an embedded layer is added into an original recognition network structure as a new plug-in. Therefore, a framework of an original identification network structure does not need to be changed, and expansion is easy. In the model application stage, the wake-up accuracy of the foundation is ensured through a first recognition result output by the foundation network and the classification layer; on the basis, a basic network and a vector distance judgment between a second recognition result output by an embedded layer and a wake-up word feature vector are added, so that whether the similarity between the current recognition result of the voice wake-up model and a preset wake-up word reaches a preset condition or not is recognized; therefore, the recognition accuracy of the whole voice wake-up model on similar voices is improved, and the false wake-up rate is effectively reduced.

The invention discloses a self-defined awakening method and device applied to intelligent voice equipment. The method comprises the steps of receiving a registration text set by a user and collectingregistration voice of the user; computing a registration awakening threshold based on the registration text; computing a registration awakening score of the registration voice based on a universal awakening model, and judging that whether the registration awakening score is more than the registration awakening threshold or not; if the registration awakening score is more than or equal to the registration awakening threshold, aligning the registration voice with the registration text, and extracting a voice segment corresponding to each character in the registration text; and extracting a Gaussian posterior feature of the voice segment corresponding to each character, and generating a feature template of the whole registration voice based on the Gaussian posterior feature of each voice segment. According to the scheme provided by the method and the device, the quality of input voice can be optimized in a registration stage, so that the system can be much finer in a score computing module, and finally, a mistakenly awakening rate of similar words also can be reduced while an awakening rate is improved.