Method for generating synthetic engine noises of an internal combustion engine in a vehicle interior and / or in an exterior area of an electric vehicle

Abstract

The invention relates to a method for generating synthetic engine noises of an internal combustion engine in a vehicle interior (100) and / or in an exterior area (101) of an electric vehicle (1), wherein a database with a plurality of interior noises (3) and / or exterior noises (4) synthetically generated by a first machine learning model, which represent interior noises and / or exterior noises of internal combustion vehicles equipped with an internal combustion engine, is provided to a computing device (8) of the electric vehicle (1), wherein, following an operator input by means of which the generation of the synthetic engine noises is activated, the computing device (8) executes a second machine learning model (7), which accesses the database and is trained thereon to perform a data mapping of input variables,which comprise a current rotational speed (6') of an electric machine of the electric vehicle and a number n ≥ 1 of vehicle-specific and driving situation-related parameters P1', P2', ..., Pn', on the associated interior noises (3) and / or exterior noises (4) of the combustion vehicle from the database, and wherein the interior noises (3) and / or exterior noises (4) are reproduced by means of a first loudspeaker system in the interior (100) of the electric vehicle (1) and / or by means of a second loudspeaker system outside the electric vehicle (1).

Description

[0001] The present invention relates to a method for generating synthetic engine noises in a vehicle interior and / or in an exterior area of an electric vehicle.

[0002] When driving an electric vehicle, which has at least one electric motor as its drive device, there is no engine noise that is significantly perceptible to the vehicle occupants, compared to a vehicle with an internal combustion engine (short: combustion vehicle). Consequently, the journey in an electric vehicle is often perceived by the vehicle occupants as quiet and not particularly emotional in terms of sound. Therefore, it may be desirable - particularly depending on the product or brand history of a vehicle manufacturer - to be able to generate engine noises that sound as authentic as possible from a vehicle with an internal combustion engine in the vehicle interior and / or in an external area of the electric vehicle during operation, in order to make the driving experience more emotional for the vehicle occupants.

[0003] To create the most authentic-sounding engine sounds from vehicles with combustion engines, a classic approach is to acoustically scan or fully sample real engine sounds from vehicles with combustion engines. However, this approach is relatively complicated, especially if the goal is to create authentic and realistic three-dimensional engine sounds in the vehicle interior and exterior of the electric vehicle.

[0004] State-of-the-art solutions already exist that utilize artificial intelligence methods to control the generation of a synthetic vehicle sound. Examples include EP 3 667 659 A1 and US 2019 / 0 392 811 A1.

[0005] DE 10 2019 111 913 A1, CN 1 09 591 693 B and DE 10 2023 001 612 A1 also deal with methods for sound design of an electric vehicle using a machine learning model.

[0006] The invention aims to provide an improved method for generating synthetic engine noises in a vehicle interior and / or in an exterior area of an electric vehicle, by means of which engine noises that sound as authentic as possible can be synthetically generated and output in the vehicle interior and / or in the exterior area of the electric vehicle.

[0007] The solution to this problem is provided by a method for generating synthetic engine noises in a vehicle interior and / or in an exterior area of an electric vehicle with the features of claim 1. The subclaims relate to advantageous developments of the invention.

[0008] According to the invention, a method is provided for generating synthetic engine noises of an internal combustion engine in a vehicle interior and / or in an exterior area of an electric vehicle, wherein a database with a plurality of interior noises and / or exterior noises synthetically generated by a first machine learning model, which represent interior noises and / or exterior noises of internal combustion vehicles equipped with an internal combustion engine, is provided to a computing device of the electric vehicle, wherein, after an operator input by means of which the generation of the synthetic engine noises is activated, a second machine learning model is executed by the computing device, which accesses the database and is trained thereon, a data mapping of input variables,which comprise a current speed of an electric machine of the electric vehicle and a number n ≥ 1 of vehicle-specific and driving situation-related parameters P1', P2', ..., Pn', on the associated interior noises and / or exterior noises of the combustion vehicle from the database, and wherein the interior noises and / or exterior noises are reproduced by means of a first loudspeaker system in the vehicle interior of the electric vehicle and / or by means of a second loudspeaker system outside the electric vehicle.

[0009] The invention provides an improved method for generating synthetic engine noises in a vehicle interior and / or in an exterior area of an electric vehicle, by means of which engine noises of a combustion engine vehicle that sound as authentic as possible can be synthetically generated in the vehicle interior and / or in the exterior area of the electric vehicle.

[0010] In one embodiment, the second machine learning model is prompted by a user input to generate interior noises of a specific vehicle model of a combustion engine vehicle and exterior noises of the same vehicle model. The interior noises and exterior noises thus originate from the same vehicle model.

[0011] In an alternative embodiment, the second machine learning model is prompted by a user input to generate interior sounds of a first vehicle model and exterior sounds of a second vehicle model, which is different from the first vehicle model. In other words, the interior sounds of a first vehicle model are combined with the exterior sounds of a second vehicle model.

[0012] In one embodiment, it is possible for the vehicle-specific and driving-situation-related parameters P1', P2', ..., Pn', which form the input variables of the second machine learning model, to include a vehicle model of the combustion engine vehicle, the type of exhaust system installed in the combustion engine vehicle, an operating mode, in particular operation in road traffic or on a racetrack, a power output, a torque, an engaged gear, preferably including the time of the gear change and the gearshift process itself, an accelerator pedal position, and a vehicle speed. At least some of the vehicle-specific and driving-situation-related parameters P1', P2', ..., Pn', such as the vehicle model of the combustion engine vehicle whose engine noise is to be synthetically generated, the type of exhaust system installed in the combustion engine vehicle, and the operating mode, can be selected by an operator input from the vehicle occupants.

[0013] In one embodiment, it is proposed that the database be generated by a further computing device outside the electric vehicle by means of a first machine learning model implemented as a Generative Adverserial Network (GAN for short).

[0014] In a preferred embodiment, a generative adverserial network is used, by means of which a complete time series with synthetic interior noises and / or exterior noises of the combustion vehicle as output variables is generated from a plurality of input variables, which include a time series with short sound samples of real interior noises and / or exterior noises of a combustion vehicle and the associated speed of a combustion engine of the combustion vehicle, and a plurality of vehicle-specific and driving situation-related parameters P1, P2, ..., Pn of the combustion vehicle, and is stored in the database. The generative adverserial network can be implemented, in particular, as a WaveGAN.

[0015] In one embodiment, it is possible for the vehicle-specific and driving situation-related parameters P1, P2, ..., Pn, which form the input variables of the Generative Adverserial Network, to include a vehicle model of the combustion vehicle, the type of exhaust system installed in the combustion vehicle, an operating mode, in particular operation in road traffic or on a race track, a power output, a torque, an engaged gear, an accelerator pedal position and a vehicle speed.

[0016] Preferably, the database can be generated by the Generative Adverserial Network in such a way that it contains complete time series with synthetic interior noises and / or exterior noises of a plurality of different vehicle models of combustion vehicles.

[0017] In an advantageous embodiment, it is proposed that a second machine learning model be used, which is implemented as a recurrent regression model. The use of a recurrent regression model has the advantage that it requires comparatively low computing resources compared to a GAN and can therefore also be executed by the computing device within the electric vehicle.

[0018] For example, a Long Short-Term Memory Network (LSTM network) can be used as a recurrent regression model.

[0019] Since the speed of the electric motor is often higher than the speed of an internal combustion engine at comparable vehicle speeds, the current speed of the electric motor can be scaled down accordingly in an advantageous embodiment. A fixed or, in an alternative embodiment, a speed-dependent scaling factor can be used for this purpose.

[0020] Further features and advantages of the present invention will become clear from the following description of a preferred embodiment with reference to the accompanying drawings. Fig. 1 a schematic representation of an electric vehicle with a first loudspeaker system and a second loudspeaker system, Fig. 2 a schematic representation of a Generative Adverserial Network, by means of which a database is generated containing synthetically generated interior noises and / or exterior noises of combustion vehicles, Fig. 3 a schematic representation of an artificial intelligence executed by a processor-based computing device of the electric vehicle.

[0021] With reference to Fig. 1, an electric vehicle 1 is shown in a schematically highly simplified form. The electric vehicle 1 has a first loudspeaker system with several loudspeakers 10a-10h in a vehicle interior 100 and with an associated subwoofer 10i in a rear area of the vehicle 1. By means of the subwoofer 10i, low frequencies (bass frequencies) can be reproduced with particular amplified effect during operation of the first loudspeaker system. Fig. The loudspeakers 10a-10h shown in Figure 1 and the associated subwoofer 10i are to be understood merely as examples with regard to their number and positions. Via the first loudspeaker system of the electric vehicle 1, which is part of an infotainment system installed in the electric vehicle 1, the vehicle interior 100 of the electric vehicle 1 can be supplied with sound, for example, from various sources selectable by the vehicle occupants via a user interface of the infotainment system. Examples of such sources, which are expressly not to be understood as exhaustive, include a DAB radio receiver, digital storage media, mobile devices that are connected to the infotainment system via a wired or wireless connection, internet streaming services, etc.

[0022] The vehicle 1 further comprises a second loudspeaker system with two external loudspeakers 11a, 11b in a front area and with two further external loudspeakers 12a, 12b in a rear area. Fig. The external loudspeakers 11a, 11b, 12a, 12b shown in Figure 1 are again to be understood as examples with regard to their number and positions. The external loudspeakers 11a, 11b, 12a, 12b can also be designed to be motor-adjustable, so that they can be dynamically pivoted about a longitudinal axis and / or a transverse axis and / or a vertical axis - comparable to a cornering light function of the front headlights of a motor vehicle.

[0023] During operation of the electric vehicle 1, which has at least one electric motor as its drive device, no significant engine noise occurs that is audibly perceptible to the vehicle occupants, compared to a vehicle with an internal combustion engine (short: combustion vehicle). Therefore, the journey in an electric vehicle is perceived by the vehicle occupants as quiet and often not particularly emotional in terms of sound.

[0024] In the following, a method for generating synthetic engine noises in the vehicle interior 100 and / or in an exterior area 101 of an electric vehicle will be explained in more detail, by means of which engine noises that sound as authentic as possible can be synthetically generated and output in the vehicle interior 100 and / or in the exterior area 101 of the electric vehicle.

[0025] The basic approach involves the use of generative, AI-based voice cloning and sampling. Furthermore, the process uses two machine learning models.

[0026] With reference to Fig.2 shows a generative adverserial network 2, which forms the first of the two machine learning models in the invention described here. This generative adverserial network 2 is designed to generate a database with complete time series of interior noises 3 and / or exterior noises 4 of combustion vehicles. For example, the generative adverserial network 2 can be a WaveGAN.

[0027] Input variables of the Generative Adverserial Network 2 form time series containing short real sound samples 5 of interior and / or exterior noises of a combustion vehicle, which are recorded with a high temporal resolution (sampling rate) during a real drive of the combustion vehicle, and the associated rotational speeds 6. This data can be collected, for example, during experimental or test drives with the combustion vehicles.

[0028] Furthermore, a number n ≥ 1 of vehicle-specific and driving-situation-related parameters P1, P2, ..., Pn are fed to the Generative Adverserial Network 2. These vehicle-specific and driving-situation-related parameters P1, P2, ..., Pn thus also form input variables of the Generative Adverserial Network 2. Possible examples of the vehicle-specific and driving-situation-related parameters P1, P2, ..., Pn, which are expressly not to be understood as exhaustive, are a vehicle model of the combustion engine vehicle, the type of exhaust system installed in the combustion engine vehicle, the power, the torque, the engaged gear, the accelerator pedal position, the vehicle speed, and the operating mode (operation in road traffic or on a racetrack).

[0029] The Generative Adverserial Network 2 is not operated in the electric vehicle 1, but by a correspondingly powerful computing device outside the electric vehicle 1. By means of the Generative Adverserial Network 2, it is possible to obtain a large database with synthetically generated, but nevertheless as realistic as possible, interior noises 3 and / or exterior noises 4 of different combustion vehicles, in particular different vehicle models of a vehicle manufacturer, and to store them in the database in a retrievable manner.

[0030] In addition to the Generative Adverserial Network 2, which forms the first of the two machine learning models used here, a further (second) machine learning model 7 is provided, which is operated by a computing device 8 installed in the electric vehicle 1. The database created by the Generative Adverserial Network 2, which contains a multitude of synthetically generated interior noises 3 and / or exterior noises 4 of different combustion vehicles, is provided to the second machine learning model 7.

[0031] Following an operator input, the generation of the synthetic engine noises is activated and the second machine learning model 7 is executed by means of the computing device 8 installed in the electric vehicle 1.

[0032] This second machine learning model 7 receives as input a time series of the current rotational speed 6' of the electric motor of the electric vehicle 1. Since the rotational speed 6' of the electric motor is often higher than the rotational speed of an internal combustion engine at comparable vehicle speeds, the current rotational speed 6' of the electric motor can advantageously be scaled down accordingly. A fixed or, in an alternative embodiment, a speed-dependent scaling factor can be used for this purpose.

[0033] Furthermore, a number n ≥ 1 of vehicle-specific and driving-situation-related parameters P1', P2', ..., Pn' are fed to the second machine learning model 7. These vehicle-specific and driving-situation-related parameters P1', P2', ..., Pn' thus also form input variables of the second machine learning model 7. Possible examples of the vehicle-specific and driving-situation-related parameters P1', P2', ..., Pn', which are expressly not to be understood as exhaustive, include the vehicle model of the combustion engine vehicle whose engine noise is to be generated, the type of exhaust system installed in the combustion engine vehicle, the operating mode (operation in road traffic or on a racetrack), the power, the torque, the engaged gear, the accelerator pedal position, and the vehicle speed. Thus, the vehicle-specific and driving-situation-related parameters P1', P2', ...'Pn', which form the input variables of the second machine learning model 7, correspond to the vehicle-specific and driving situation-related parameters P1, P2, ..., Pn of the Generative Adverserial Network 2. However, this is not mandatory.

[0034] At least one of the vehicle-specific and driving situation-related parameters P1', P2', ..., Pn', in particular the vehicle model whose engine noise is to be synthetically generated, can be selected by an operator input from the vehicle occupants.

[0035] The second machine learning model 7 can, in particular, be implemented as a low-computationally intensive regression model, so that during productive operation within the electric vehicle 1, it requires little computing power from the computing device 8 installed in the electric vehicle 1. The second machine learning model 7 is preferably implemented as a recurrent neural network, in particular as a long short-term memory network (LSTM network for short).

[0036] The second machine learning model 7 is trained to perform a data mapping of the input variables, i.e., the current rotational speed 6' of the electric motor and the vehicle-specific and driving situation-related parameters P1', P2', ..., Pn', to the corresponding interior noises 3 and / or exterior noises 4 and thus to extract them from the provided database. These sound signals 3, 4 are output using the first loudspeaker system in the vehicle interior 100 and / or using the second loudspeaker system in the exterior area 101 of the vehicle 1. As mentioned above, the first loudspeaker system is the loudspeaker system of an infotainment system installed in the electric vehicle 1. In an alternative embodiment, it can be provided that the first loudspeaker system is intended only for outputting the interior noises 3, and that the infotainment system has its own loudspeaker system.

[0037] The training of the second machine learning model 7 can be carried out in particular using the database generated by the Generative Adverserial Network 2.

[0038] The second machine learning model 7 can be caused by a user input to generate interior sounds 3 of a vehicle model and exterior sounds 4 of the same vehicle model.

[0039] Preferably, it is also possible for the vehicle occupants to combine the interior noise 3 of a vehicle model of a first series (“Series A”) with the exterior noise 4 of a vehicle model of a second series (“Series B”), which is different from the first series, by means of corresponding operating inputs.

Claims

[1] Method for generating synthetic engine noises of an internal combustion engine in a vehicle interior (100) and / or in an exterior area (101) of an electric vehicle (1), wherein a database with a plurality of interior noises (3) and / or exterior noises (4) synthetically generated by a first machine learning model, which represent interior noises (3) and / or exterior noises (4) of internal combustion vehicles equipped with an internal combustion engine, is provided to a computing device (8) of the electric vehicle (1), wherein, following an operator input by means of which the generation of the synthetic engine noises is activated, the computing device (8) executes a second machine learning model (7), which accesses the database and is trained thereon, a data mapping of input variables,which comprise a current rotational speed (6') of an electric machine of the electric vehicle (1) and a number n ≥ 1 of vehicle-specific and driving situation-related parameters P1', P2', ..., Pn', on the associated interior noises (3) and / or exterior noises (4) of the combustion vehicle from the database, and wherein the interior noises (3) and / or exterior noises (4) are reproduced by means of a first loudspeaker system in the vehicle interior (100) of the electric vehicle (1) and / or by means of a second loudspeaker system outside the electric vehicle (1). [2] Method according to claim 1, characterized by that the second machine learning model (7) is caused by a user input to generate interior noises (3) of a specific vehicle model of a combustion engine vehicle and exterior noises (4) of the same vehicle model. [3] Method according to claim 1, characterized bythat the second machine learning model (7) is caused by a user input to generate interior noises (3) of a first vehicle model of a combustion vehicle and exterior noises (4) of a second vehicle model of a combustion vehicle, which is different from the first vehicle model. [4] Method according to one of claims 1 to 3, characterized by that the vehicle-specific and driving situation-related parameters P1', P2', ..., Pn', which form the input variables of the second machine learning model (7), comprise a vehicle model of the combustion vehicle, the type of exhaust system installed in the combustion vehicle, an operating mode, in particular operation in road traffic or on a race track, a power output, a torque, an engaged gear, an accelerator pedal position and a vehicle speed. [5] Method according to one of claims 1 to 4, characterized bythat the database is generated by a further computing device outside the electric vehicle (1) by means of a first machine learning model designed as a generative adversarial network (2). [6] Method according to claim 5, characterized by that a generative adversarial network (2) is used, by means of which a complete time series with synthetic interior noises (3) and / or exterior noises (4) of the combustion vehicle as output variables is generated from a plurality of input variables, which comprise a time series with short sound samples (5) of real interior noises and / or exterior noises of the combustion vehicle and the associated speed of the combustion engine of the combustion vehicle and a plurality of vehicle-specific and driving situation-related parameters P1, P2, ..., Pn of the combustion vehicle, and is stored in the database. [7] Method according to claim 6, characterized bythat the vehicle-specific and driving situation-related parameters P1, P2, ..., Pn, which form the input variables of the Generative Adversarial Network (2), include a vehicle model of the combustion vehicle, the type of exhaust system installed in the combustion vehicle, an operating mode, in particular operation in road traffic or on a race track, a power output, a torque, an engaged gear, an accelerator pedal position and a vehicle speed. [8] Method according to one of claims 6 or 7, characterized by that the database is generated by the Generative Adversarial Network (2) in such a way that it contains complete time series with synthetic interior noises (3) and / or exterior noises (4) of a plurality of different vehicle models of combustion vehicles. [9] Method according to one of claims 1 to 8, characterized bythat a second machine learning model (7) is used, which is designed as a recurrent regression model, in particular as a long short-term memory network. [10] Method according to one of claims 1 to 9, characterized by that the current speed (6') of the electric machine is scaled down.

Images