Systems, methods, and devices for wakeup word detection with continuous learning
Systems, methods, and devices detect audio signals. Methods may include receiving an audio input at an audio front end circuit, and identifying, using a first circuit, a wake word based, at least in part, on the received audio signal. Methods may also include determining if the identifying performed by the first circuit should be modified based, at least in part, on a designated threshold of accuracy associated with the first circuit and a result of the identifying, the designated threshold of accuracy being determined based, at least in part, on a language processing model of a second circuit. Moreover, the modifying may include adjusting wake word detection operations of the first circuit based, at least in part, on a result of the determining.
1 . A method comprising:
receiving an audio input at an audio front end circuit;
identifying, using a first circuit, a wake word based, at least in part, on the received audio input;
mapping, using a language processing model, a distribution of decisions made by a second circuit in a latent space of the second circuit to a latent space of the first circuit being a low-dimensional space, and the audio input comprising audio data in an observable space;
generating a designated threshold of accuracy based on a distance threshold between a positive distribution of decisions and a negative distribution of decisions made by the first circuit in the latent space of the first circuit and representing correct and incorrect decisions made by the first circuit as determined relative to the distribution of decisions made by the second circuit after being mapped to the low-dimensional latent space of the first circuit;
determining if the identifying performed by the first circuit should be adjusted based, at least in part, on the designated threshold of accuracy, and
adjusting, using one or more processors, wake word detection operations of the first circuit responsive to determining that the distance between the distributions is greater than the designated threshold of accuracy.
2 . The method of claim 1 , wherein the first circuit operates at a first power level, and the second circuit operates at a second power level.
3 . The method of claim 2 , wherein the second power level is higher than the first power level.
4 . The method of claim 1 , wherein the designated threshold of accuracy represents a designated percentage of correct wake word identifications.
5 . The method of claim 4 , wherein the designated threshold of accuracy is determined based, at least in part, on a plurality of statistical distributions in a latent space associated with the first circuit.
6 . The method of claim 5 , wherein the designated threshold of accuracy is determined based, at least in part, on a distance between the first distribution and the second distribution identifying a separation between the first distribution and the second distribution.
7 . The method of claim 1 , wherein the language processing model represents the second circuit in a latent space associated with the first circuit.
8 . The method of claim 7 , wherein the language processing model is updated based on one or more wake word identification operations of the second circuit.
9 . The method of claim 1 , wherein the modifying further comprises:
adjusting one or more weights associated with a wake word identification operation of the first circuit.
10 . A system comprising:
an audio front end circuit configured to receive an audio input;
a first circuit comprising one or more processors configured to identify a wake word based, at least in part, on the received audio input; and
a learning module comprising one or more processors configured to:
map a distribution of decisions made by a second circuit in a latent space of the second circuit to a latent space of the first circuit being a low-dimensional space, and the audio input comprising audio data in an observable space;
generate a designated threshold of accuracy based on a distance threshold between a positive distribution of decisions and a negative distribution of decisions made by the first circuit in the latent space of the first circuit and representing correct and incorrect decisions made by the first circuit as determined relative to the distribution of decisions made by the second circuit after being mapped to the low-dimensional latent space of the first circuit; and
determine that the identifying performed by the first circuit should be adjusted based, at least in part, on the designated threshold of accuracy,
the one or more processors being further configured to adjust wake word detection operations of the first circuit responsive to determining that the distance between the distributions is greater than the designated threshold of accuracy.
11 . The system of claim 10 , wherein the first circuit operates at a first power level, and the second circuit operates at a second power level, and wherein the second power level is higher than the first power level.
12 . The system of claim 10 , wherein the designated threshold of accuracy represents a designated percentage of correct wake word identifications.
13 . The system of claim 12 , wherein the designated threshold of accuracy is determined based, at least in part, on a plurality of statistical distributions in a latent space associated with the first circuit.
14 . The system of claim 13 , wherein the designated threshold of accuracy is determined based, at least in part, on a distance between the first distribution and the second distribution identifying a separation between the first distribution and the second distribution.
15 . The system of claim 10 , wherein the language processing model represents the second circuit in a latent space associated with the first circuit, and wherein the language processing model is updated based on one or more wake word identification operations of the second circuit.
16 . A device comprising:
a first circuit configured to identify a wake word based, at least in part, on a received audio input; and
a learning module comprising one or more processors configured to:
map a distribution of decisions made by a second circuit in a latent space of the second circuit to a latent space of the first circuit being a low-dimensional space, and the audio input comprising audio data in an observable space;
generate a designated threshold of accuracy based on a distance threshold between a positive distribution of decisions and a negative distribution of decisions made by the first circuit in the latent space of the first circuit and representing correct and incorrect decisions made by the first circuit as determined relative to the distribution of decisions made by the second circuit after being mapped to the low-dimensional latent space of the first circuit; and
determine if the identifying performed by the first circuit should be adjusted based, at least in part, on the designated threshold of accuracy,
the one or more processors being further configured to adjust wake word detection operations of the first circuit responsive to determining that the distance between the distributions is greater than the designated threshold of accuracy.
17 . The device of claim 16 , wherein the first circuit operates at a first power level, and the second circuit operates at a second power level, and wherein the second power level is higher than the first power level.
18 . The device of claim 16 , wherein the designated threshold of accuracy represents a designated percentage of correct wake word identifications.
19 . The device of claim 18 , wherein the designated threshold of accuracy is determined based, at least in part, on a plurality of statistical distributions in a latent space associated with the first circuit, and wherein the designated threshold of accuracy is determined based, at least in part, on a distance between the first distribution and the second distribution identifying a separation between the first distribution and the second distribution.
20 . The device of claim 16 , wherein the language processing model represents the second circuit in a latent space associated with the first circuit, and wherein the language processing model is updated based on one or more wake word identification operations of the second circuit.