Vehicle language processing
A computing system can be programmed to receive a spoken language command in response to emitting a spoken language cue and process the spoken language command with a generalized adversarial neural network (GAN) to determine a vehicle command. The computing system can be further programmed to operate a vehicle based on the vehicle command.
1. A method, comprising:
receiving a spoken language command in response to emitting a spoken language cue;
training a neural network (NN) at least in part by inputting mel-frequency data from the spoken language command combined with a loss function to a convolutional filter bank that outputs modified mel-frequency data to a bidirectional long short-term memory that in turn outputs audio spectrum data to determine a vehicle command using a generalized adversarial neural network (GAN) that determines a loss function by a combination of a binary classification of the output audio spectrum data as one of real or fake and a ground truth-based loss;
processing the spoken language command with the NN to determine the vehicle command; and
operating a vehicle based on the vehicle command.
2. The method of claim 1 , further comprising transforming the spoken language command into mel frequency samples before processing with the NN.
3. The method of claim 2 , wherein the mel frequency samples are each compressed into a single vector by convolving mel values along a y-axis before processing with the NN.
4. The method of claim 3 , wherein a mel frequency scale is a log power spectrum of spoken language command frequencies on a nonlinear scale of frequencies.
5. The method of claim 1 , further comprising training the GAN to determine real or fake spoken language using a plurality of recorded spoken language commands, ground truth that identifies the recorded spoken language commands as real or fake and a loss function based on ground truth.
6. The method of claim 1 , wherein operating the vehicle includes determining a path polynomial based on the vehicle command.
7. The method of claim 1 , wherein operating the vehicle includes determining a cognitive map based on vehicle sensor data.
8. The method of claim 1 , further comprising processing synthetic language data with a GAN to determine the spoken language cue.
9. The method of claim 1 , wherein the vehicle command is a request for goal-directed behavior of the vehicle.
10. A system, comprising a processor; and
a memory, programmed to:
receive a spoken language command in response to emitting a spoken language cue;
train a neural network (NN) at least in part by inputting mel-frequency data from the spoken language command combined with a loss function to a convolutional filter bank that outputs modified mel-frequency data to a bidirectional long short-term memory that in turn outputs audio spectrum data to determine a vehicle command using a generalized adversarial neural network (GAN) that determines a loss function by a combination of a binary classification of the output audio spectrum data as one of real or fake and a ground truth-based loss;
process the spoken language command with the NN to determine the vehicle command; and
operate a vehicle based on the vehicle command.
11. The system of claim 10 , further comprising transforming the spoken language command into mel frequency samples before processing with the NN.
12. The system of claim 11 , wherein the mel frequency samples are each compressed into a single vector by convolving mel values along a y-axis before processing with the NN.
13. The system of claim 12 , wherein a mel frequency scale is a log power spectrum of spoken language command frequencies on a nonlinear scale of frequencies.
14. The system of claim 10 , further comprising training the GAN to determine real or fake spoken language using a plurality of recorded spoken language commands, ground truth that identifies the recorded spoken language commands as real or fake and a loss function based on ground truth.
15. The system of claim 10 , wherein operating the vehicle includes determining a path polynomial based on the vehicle command.
16. The system of claim 10 , wherein operating the vehicle includes determining a cognitive map based on vehicle sensor data.
17. The system of claim 10 , further comprising processing synthetic language data with a GAN to determine the spoken language cue.
18. The system of claim 10 , wherein the vehicle command is a request for goal-directed behavior of the vehicle.
19. A system, comprising:
means for controlling second vehicle steering, braking and powertrain;
computer means for:
receiving a spoken language command in response to emitting a spoken language cue;
training a neural network (NN) at least in part by inputting mel-frequency data from the spoken language command combined with a loss function to a convolutional filter bank that outputs modified mel-frequency data to a bidirectional long short-term memory that in turn outputs audio spectrum data to determine a vehicle command using a generalized adversarial neural network (GAN) that determines a loss function by a combination of a binary classification of the output audio spectrum data as one of real or fake and a ground truth-based loss;
processing the spoken language command with the NN to determine the vehicle command; and
operating a vehicle based on the vehicle command and the means for controlling vehicle steering, braking and powertrain.
20. The system of claim 19 , further comprising transforming the spoken language command into a mel frequency samples before processing with the NN.