Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings. These embodiments are exemplary and not intended to limit the scope of the present invention.
FIG. 1 is a diagram illustrating a speech recognition method according to an embodiment of the present invention. As shown in fig. 1, according to an embodiment of the present invention, a noise frequency pre-determination is first performed in step S100. Fig. 2 shows a schematic flow diagram of noise pre-determination according to an embodiment of the invention.
As shown in fig. 2, according to one embodiment, first, noise frequency prediction key information is collected in step S101. In one embodiment, the noise frequency prediction key information is current vehicle speed information of the vehicle. Various approaches may be taken to collect the current vehicle speed information of the host vehicle. As is well known, the prior art has various methods for collecting the speed of the vehicle. The vehicle speed information can be displayed on the dashboard of the vehicle. The vehicle speed information may be obtained by measurement of a speed sensor or the like, or may be obtained by the rotation speed of the wheels, and the vehicle speed information obtained anyway is within the scope of the present invention. In another embodiment, the noise frequency prediction key information is both the current vehicle speed information and the link information of the host vehicle. The current vehicle speed information is collected as described above, and is not described herein again. The current road section information may be known from the current position information of the vehicle. The current position information of the vehicle can be obtained by technologies such as GPS positioning and beidou positioning, and can be compared with a prestored map, so as to obtain current link information, where the link information can be the name of a road, road section (for example, S305 lane, G210 national daoza section 1898 km to 1908 km, shenhai highway section), and road section end (for example, a link between longitude 120.24320115063475 latitude 30.217615576158664-longitude 120.24359811756895 latitude 30.20504097488256) divided according to the information obtained by GPS information or beidou positioning technology. In a further embodiment, the noise frequency anticipation key information further includes window switch information. In this case, the vehicle may collect window opening and closing information. When the window is opened, the wind noise amplitude and frequency can change. The window opening and closing information may include a combined opening and closing condition of each window. For example, the case where the front left window is open and the remaining windows are closed; the condition that the front right window is opened and the other windows are closed; all windows open or all windows closed, etc. The window opening and closing information may include information on whether the window is half-opened or whether the window has a gap. In this case, local wind speed information can be acquired. For example, a wind speed sensor may be provided in the vehicle cabin to detect the wind speed and wind direction in the vehicle cabin. The wind speed sensor can be arranged outside the carriage to detect the wind speed and the wind direction outside the carriage. The wind speed information may be only the vehicle speed and/or the wind direction inside the vehicle compartment, may be only the vehicle speed and/or the wind direction outside the vehicle compartment, or may be both of them.
Then, in step S102, the noise frequency is predicted based on the collected noise frequency prediction key information. The collected noise frequency pre-determination key information may be compared with a pre-stored noise frequency comparison table to pre-determine the noise frequency. The noise frequency comparison chart can establish the corresponding relation between the noise frequency pre-judgment key information and the noise frequency, so that the noise frequency can be determined according to the collected noise frequency pre-judgment key information through a searching method.
According to one embodiment, the noise frequency comparison chart is stored in the vehicle, and may be updated periodically or aperiodically by the cloud device, or periodically or aperiodically by the driver or the vehicle owner. According to one embodiment, the noise frequency comparison graph is created from big data, i.e. data collected from a plurality of vehicles of the same type. The cloud device refers to a remote server and may include one or more computers or processors. The processors may be centralized or distributed, and together may perform the functions of embodiments of the invention. In this context, the same type of vehicle may be used with a relatively relaxed definition. For example, in one embodiment, the same manufacturer, same engine, same displacement, same configuration of automobiles is referred to. In one embodiment, it may refer to the same engine, the same displacement car. The smaller the number of vehicles on the market, the larger the range of definition.
According to one embodiment, a noise frequency comparison graph may be generated from big data by the cloud device according to the following method. First, the cloud device collects noise information from vehicles of the same type as the vehicle. In the process of driving of the vehicles of the same type, a microphone MIC for reporting noise outputs received sound signals (including tire noise, wind noise, voice, music sound and the like) to a vehicle processor, a vehicle-mounted spectrum analysis device (such as the vehicle processor) performs spectrum analysis on the sound signals, the result of the spectrum analysis is the relationship between the frequency points of the sound signals and the energy of the frequency points, the duration time of the signals of the frequency points is judged, if the duration time exceeds a certain threshold value, the sound signals are regarded as continuous single signals, namely, the sound signals are noise signals, several noise signals with higher energy are screened, the frequencies of the noise signals are recorded, and the frequencies and vehicle speed signals and/or GPS positioning information (examples of road section information) corresponding to the frequencies are sent to a cloud device together. Those skilled in the art will appreciate that more information, such as window status information, may be collected and sent to the cloud. These relatively high energy signals collected may include, for example, wind noise, tire noise, and the like. In another embodiment, the vehicles may send the sound signals collected by the microphones directly to the cloud device, which performs the spectral analysis. Then, the cloud device performs statistical analysis on collected data reported by a large number of vehicles of the vehicle type, and because the noise frequencies of most single vehicle types on the same road surface and at the same vehicle speed are very close, the noise frequencies of each road section and at various vehicle speeds can be obtained, the corresponding relation between the noise frequencies and the road sections and the vehicle speeds is formed, and a noise frequency comparison table (data table) is compiled to express the mapping relation. A vehicle model has a data table with two dimensions, such as the abscissa filling the road segment (GPS positioning information), the ordinate filling the vehicle speed (such as 80KM/H), and then the intersection filling the noise frequency of the vehicle speed of the road segment, such as 180 HZ. Table 1 gives an example of such a graph.
TABLE 1
It should be understood by those skilled in the art that the comparison table of the vehicle speed and the noise frequency may be made only, or the comparison table of the road section and the noise frequency may be made only, that is, the corresponding comparison table may be made according to the noise frequency pre-judgment key information that can be collected by the vehicle.
It will be appreciated by those skilled in the art that the noise contrast chart may be in the form of not only a data table, but also a data map, such as a graph.
Returning to fig. 1, after the noise frequency is pre-determined, real-time noise is detected according to the pre-determined frequency at step S200. This can be achieved, for example, with a microphone for noise monitoring. The noise monitor microphone may be implemented by the above noise report microphone. The noise monitoring microphones may be plural and are respectively provided in the vicinity of the noise source. For example, the noise monitoring microphone may include a tire noise monitoring microphone provided at a position of the vehicle frame corresponding to the wheel. For another example, the noise monitor microphone may include an engine monitor microphone, which is disposed in the vicinity of the engine. The noise monitoring microphone may further include a microphone provided in the vehicle cabin for monitoring wind noise, and may further include a microphone for monitoring current noise. The current noise is generally noise generated by PWM brightness adjustment or a switching power supply, and the current noise of the same type of vehicle is basically the same. The frequency and the power of the current noise can be obtained by performing spectrum analysis on the signal received by the microphone in the quiet state of the vehicle.
The noise monitoring microphone may be provided with a noise collection filter that is set according to the pre-determined noise frequency so as to collect only real-time noise in the vicinity of the pre-determined noise frequency. The vicinity range is, for example, a frequency range determined by increasing or decreasing a predetermined frequency value (e.g., increasing or decreasing 50 Hz). The parameters of the noise collecting filter can be adjusted in real time, so that the parameters of the noise collecting filter can be adjusted according to the pre-judged noise frequency, real-time noise can be collected more accurately, and the noise collecting filter can be well adapted to the change of vehicle conditions and road conditions.
Next, in step S300, a frequency corresponding to noise having the largest real-time noise energy is determined from the real-time noise. It should be noted that in this context, energy maximum means that the noise energy is maximum in the vicinity. If a plurality of noise frequencies are determined in step S100, then a corresponding plurality of the nearby ranges, and thus a corresponding plurality of noise frequencies with the largest energy, will exist in step S200.
When the frequency with the largest real-time noise energy is determined, the signal can be subjected to spectrum analysis by using fast fourier transform, so that the frequencies contained in the signal and the power corresponding to each frequency are obtained. The frequency with the maximum power is selected.
Finally, in step S400, the signal of the frequency at which the noise energy is the largest is removed or attenuated from the picked-up or collected voice through a voice filter of a voice pickup device (e.g., a microphone). At this time, a band-stop filter may be provided. For example, if the noise is determined to be 350HZ as above, the cutoff frequency of the band stop filter is set to 350HZ, and the noise frequency is not allowed to pass.
The voice pickup device is disposed at a position where voice is conveniently collected. The voice recognition device is provided, for example, on a panel in front of the driver for collecting the voice of the driver, and may be provided, for example, on the back of a rear seat for collecting the voice of a rear person.
The filter can adopt a Chebyshev filter, a Butterworth filter and the like, and the car machine can adjust the parameters of the filter according to the frequency with the maximum real-time noise energy, so that the filter can adapt to different noise frequencies in real time. The filter can be adjusted in real time so that it can adapt well to changes in vehicle conditions and road conditions.
Such as when the vehicle is traveling, the processor performs real-time noise detection in the range of 400HZ to 500HZ in step S200 by predicting that the noise frequency is 450HZ by looking up in the correspondence table in step S100, and then performs corresponding attenuation, such as attenuation of 10db, on the frequency of about 455HZ in the received MIC signal if the frequency noise frequency with the largest energy is determined to be 455HZ in S300. The method can combine the big data of the vehicle type with the actual situation of the vehicle, has fast calculation, can improve the signal-to-noise ratio of the MIC without additional or expensive hardware, and has low cost.
In addition, the car machine can identify the voice signal after noise attenuation, thereby greatly improving the voice identification rate.
FIG. 3 is a schematic diagram of a speech recognition system according to one embodiment of the present invention.
As shown in fig. 3, a speech recognition system 10 according to an embodiment of the present invention includes: a noise frequency prejudging device 100 that prejudges a noise frequency that may appear in the collected speech; a real-time noise detection device 200 for detecting real-time noise from the pre-determined noise frequency; an energy maximum noise frequency determination means 300 that determines a frequency at which the energy of the real-time noise is maximum from the real-time noise; and a speech detection attenuation device 400 that removes or attenuates the signal of the frequency at which the real-time noise energy is the largest from the picked-up or collected speech.
According to one embodiment, the noise frequency prejudging device 100 performs noise prejudging as follows: collecting noise frequency pre-judgment key information; and determining the noise frequency by searching a noise frequency comparison chart according to the noise frequency pre-judging key information, wherein the device is used for the vehicle, and the noise frequency comparison chart is made by collecting the correlation data of the noise of the vehicle of the same vehicle type and the noise frequency pre-judging key information.
According to one embodiment, the speech recognition system 10 further comprises a noise frequency comparison chart updating unit (not shown), which may be configured to update the noise frequency comparison chart periodically or aperiodically. According to one embodiment, the update is performed by receiving a new noise frequency comparison graph from the cloud device periodically or aperiodically.
According to one embodiment, the real-time noise detection device 200 detects real-time noise by a noise detection device disposed near a noise source, the noise detection device being provided with a filter, the parameters of which are set according to the noise frequency, so that only noise within a certain range of the noise frequency is collected. The filter may be a band pass filter which may be different from the attenuation means which removes noise from the speech, the attenuation means being substantially a band stop filter.
The invention also relates to a vehicle comprising the system or adopting the method.
Those skilled in the art will appreciate that the above devices may be implemented by special hardware, such as a field programmable gate array, a single chip, or a microchip, or by a combination of software and hardware.
The description of the method of the present invention may be used for understanding the description of the system, and the description of the system may also be used for understanding the method of the present invention.
The above description is intended to be illustrative, and not restrictive, and any changes and substitutions that come within the spirit of the invention are desired to be protected.