Method and system for recognizing a user utterance

1. A method of generating textual representations of a user utterance, the user utterance being collected by an electronic device associated with a user, the user utterance being provided in response to machine-generated utterances outputted by the electronic device, the electronic device being configured to be communicatively coupled with a server, the server for generating the machine-generated utterances, the method executable by the server, the method comprising:

acquiring, by the server from the electronic device, an audio signal being an audio representation of the user utterance,

the user utterance being in response to a given machine-generated utterance previously outputted by the electronic device to the user;

acquiring, by the server, a machine-generated text string being a textual representation of the given machine-generated utterance; and

generating, by the server using a Speech-to-Text (STT) model, another text string based on the audio signal and the machine-generated text string,

the another text string being a textual representation of the user utterance while taking into account the machine-generated text string as a context of the user utterance.

2. The method of claim 1 , wherein the method further comprises:

generating, by the server using the STT model, a first text string based on a first audio signal,

the first audio signal being an audio representation of a first user utterance and the first text string being a textual representation of the first user utterance;

generating, by the server using another model, the machine-generated text string based on the first text string; and

generating, by the server using a Text-to-Speech (TTS) model, a machine-generated audio signal based on the machine-generated text string,

the machine-generated audio signal being an audio representation of the given machine-generated utterance.

3. The method of claim 1 , wherein the STT model is a neural network (NN) having an encode-decoder architecture including an encoder stack of layers and a decoder stack of layers, and wherein the generating the another text string comprises:

generating, by the server, using an acoustic embedding algorithm, an audio vector representative of the audio signal;

inputting, by the server, the audio vector to the encoder stack of the NN;

generating, by the server, using a text embedding algorithm, a text vector representative of the machine-generated text string; and

inputting, by the server, the text vector to the decoder stack of the NN.

4. The method of claim 3 , wherein the NN is a Transformer-based NN.

5. The method of claim 1 , wherein the method further comprises, during a training phase, training the STT model based on a training set of data, the training set of data comprising a plurality of training objects, a given training object including: (i) an indication of a training audio signal generated based on a training user utterance uttered by a training user; (ii) a first training text string, the first training text string being a textual representation of the training user utterance; and (iii) a second training text string, the second training text string being a textual representation of a respective machine-generated utterance in response to which the training user has uttered the training user utterance, the second training text string used for providing a context for the first training text string.

6. The method of claim 2 , wherein the method further comprises acquiring, by the server, the first audio signal from the electronic device.

7. The method of claim 2 , wherein the method further comprises transmitting, by the server, the machine-generated audio signal to the electronic device.

8. The method of claim 2 , wherein the another model includes a Natural Language Processing (NLP) model.

9. The method of claim 2 , wherein the STT model, the TTS model, and the another model are hosted by another server.

10. The method of claim 8 , wherein the NLP model is a Transformer-based NN model.

11. The method of claim 2 , wherein the TTS model is a Transformer-based NN model.

12. The method of claim 2 , wherein the method further comprises:

generating, by the server using the STT model, a second text string based on a second audio signal,

the second audio signal being an audio representation of a second user utterance, the second user utterance being a follow-up request to the given machine-generated utterance;

generating, by the server using the another model, another machine-generated text string based on the second text string,

the another machine-generated text string being a textual representation of another machine-generated utterance to be provided to the user;

generating, by the server using the TTS model, another machine-generated audio signal based on the another machine-generated text string,

the another machine-generated audio signal being an audio representation of the another machine-generated utterance to be provided to the user in response to the second user utterance;

acquiring, by the server, a third audio signal from the electronic device,

the third audio signal being an audio representation of a third user utterance, the third user utterance being another follow-up request to the another machine-generated utterance; and

generating, by the server using the STT model, a third text string based on the third audio signal, the machine-generated text string and the another machine-generated text string,

the third text string being a textual representation of the third user utterance while taking into account the machine-generated text string and the another machine-generated text string as the context of the another follow-up request.

13. The method of claim 12 , wherein the method further comprises transmitting, by the server, the another machine-generated audio signal to the electronic device.

14. A server for generating textual representations of a user utterance, the user utterance being collected by an electronic device, communicatively coupled with the server, associated with a user, the user utterance being provided in response to machine-generated utterances outputted by the electronic device, the server comprising:

a processor;

a non-transitory computer-readable medium storing instructions; and

the processor, upon executing the instructions, being configured to:

acquire, from the electronic device, an audio signal being an audio representation of the user utterance,

the user utterance being in response to a given machine-generated utterance previously outputted by the electronic device to the user;

acquire, a machine-generated text string being a textual representation of the given machine-generated utterance; and

generate, using a Speech-to-Text (STT) model, another text string based on the audio signal and the machine-generated text string,

the another text string being a textual representation of the user utterance while taking into account the machine-generated text string as a context of the user utterance.

15. The server of claim 14 , wherein the processor is further configured to:

generate, using the STT model, a first text string based on a first audio signal,

the first audio signal being an audio representation of a first user utterance and the first text string being a textual representation of the first user utterance;

generate, using another model, the machine-generated text string based on the first text string; and

generate, using a Text-to-Speech (TTS) model, a machine-generated audio signal based on the machine-generated text string,

the machine-generated audio signal being an audio representation of the given machine-generated utterance.

16. The server of claim 14 , wherein the STT model is a neural network (NN) having an encode-decoder architecture including an encoder stack of layers and a decoder stack of layers, and wherein the processor is configured to generate the another text string by:

generating, using an acoustic embedding algorithm, an audio vector representative of the audio signal;

inputting the audio vector to the encoder stack of the NN;

generating, using a text embedding algorithm, a text vector representative of the machine-generated text string; and

inputting, the text vector to the decoder stack of the NN.

17. The server of claim 16 , wherein the NN is a Transformer-based NN.

18. The server of claim 14 , wherein the processor is further configured, during a training phase, to train the STT model based on a training set of data, the training set of data comprising a plurality of training objects, a given training object including: (i) an indication of a training audio signal generated based on a training user utterance uttered by a training user; (ii) a first training text string, the first training text string being a textual representation of the training user utterance; and (iii) a second training text string, the second training text string being a textual representation of a respective machine-generated utterance in response to which the training user has uttered the training user utterance, the second training text string used for providing a context for the first training text string.

19. The server of claim 15 , wherein the another model includes a Natural Language Processing (NLP) model.

20. The server of claim 15 , wherein the STT model, the TTS model, and the another model are hosted by another server to which the server is communicatively coupled.