Conversation facilitating method and electronic device using the same

1. A computer-implemented conversation facilitating method, comprising:

providing a robot comprises a camera, a microphone, a speaker and a processor, wherein the camera, the microphone and the speaker are electrically connected to the processor;

obtaining, through the camera, one or more images of a plurality of participants of a conversation;

identifying, through the processor, the participants based on the one or more images;

localizing, through the processor, relative positions of the participants based on the one or more images;

detecting, through the microphone, a plurality of speeches of the conversation;

matching, through the processor, one of the participants to each of the detected speeches according to the relative positions of the participants and counting participations of the matched participant in the conversation;

identifying, through the processor, a passive subject from all the participants according to the participations of all the participants in the conversation;

finding, through the processor, a topic of the conversation between the participants; and

controlling, through the processor, the robot to move to align with the passive subject according to the relative position of the passive subject, and addressing, through the speaker, the passive subject and speaking a sentence related to the topic to engage the passive subject;

wherein the conversation comprises a plurality of turns of dialogue; the counting, through the processor, the participations of the participant in the conversation comprises:

counting, through the processor, a total speaking time of the matched participant during a last αN turns of dialogue in the conversation, wherein α is an integer greater than 0, N is a total number of the participations; and

the conversation comprises a plurality of turns of dialogue; the identifying, through the processor, the passive subject from all the participants according to the participations of all the participants in the conversation comprises:

identifying, through the processor, the participant with the total speaking time below 2 standard deviations of an average speaking time in the conversation during the last αN turns of dialogue as the passive subject.

2. The method of claim 1 , wherein the finding, through the processor, the topic of the conversation between the participants comprises:

transcribing, through the processor, each utterance in the speeches of the conversation into a text;

encoding, through the processor, the text corresponding to one or more spoken sentences in the conversation into a sentence-level semantic vector, wherein the sentence-level semantic vector is an embedding or dense vector and represents a meaning of a word or sentence; and

selecting, through the processor, a fact with a closest similarity to the sentence-level semantic vector from a pre-selected fact collection, wherein the similarity meets a threshold; and

the speaking the sentence related to the topic through the speaker comprises:

speaking the sentence comprising the fact through the speaker.

3. The method of claim 2 , wherein the fact is a trivia related to the topic, and the pre-selected fact collection is a corpus of topical facts comprising a plurality of single-sentence entries; and the selecting, through the processor, the fact with the closest similarity to the sentence-level semantic vector from the pre-selected fact collection comprises:

comparing, through the processor, a vector of each single-sentence entry in the corpus of topical facts with the sentence-level semantic vector; and

taking, through the processor, the single-sentence entry of the vector with a closest cosine similarity to the sentence-level semantic vector as the trivia; and

the speaking the sentence related to the topic through the speaker comprises:

speaking the sentence comprising the trivia through the speaker.

4. The method of claim 2 , wherein the microphone is a microphone array comprising a plurality of microphones disposed on a head of the robot in a circular fashion;

the detecting, through the microphone, the speeches of the conversation comprises:

receiving, through the microphone array, an audio; and

detecting, through the processor, the speeches of the conversation from the audio using voice activity detection to send a detected audio to an automatic speech recognition engine;

the transcribing each utterance in the speeches of the conversation into the text comprises:

transcribing, through the automatic speech recognition engine, each utterance in the speeches of the conversation into the text based on the detected audio; and

the matching, through the processor, one of the participants to each of the detected speeches according to the relative positions of the participants comprises:

matching, through the processor, each of the participants with a direction of arrival based on a number and the relative positions of the participants, wherein the direction of arrival represents an angle of a voice of each of the participants to arrive the microphone, and a range of the direction of arrival is between 0 and 180 degrees.

5. The method of claim 2 , wherein the speaking the sentence comprising the fact through the speaker comprises:

formatting, through the processor, the sentence comprising the fact using natural language processing; and

converting, through the processor, the sentence into speech using a text-to-speech convertor, and speaking, through the speaker, the sentence comprising the fact; and

wherein the robot further comprises a screen, and the method further comprises:

showing, through the screen, facial expressions to the passive subject.

6. The method of claim 1 , wherein the counting, through the processor, the participations of the participant in the conversation further comprises:

counting, through the processor, a number of participation of the matched participant during the last αN turns of dialogue in the conversation; and

the identifying, through the processor, the passive subject from all the participants according to the participations of all the participants in the conversation further comprises:

identifying, through the processor, the participant with the number of participation less than β times during the last αN turns of dialogue as the passive subject, wherein β is an integer smaller than α and greater than 0.

7. The method of claim 1 , further comprising:

navigating, through the processor, the robot to move and avoid obstacles in an environment of the conversation according to a planned trajectory during the conversation to detect the participants in a dynamic manner; and

in response to a change in the topic or the participants, returning to perform the step of obtaining, through the camera, the one or more images of the plurality of participants of the conversation.

8. An electronic device, comprising:

a camera;

a microphone;

a speaker;

one or more processors; and

one or more memories storing one or more programs configured to be executed by the one or more processors, wherein the electronic device is a robot, and the one or more programs comprise instructions to:

obtain, through the camera, one or more images of a plurality of participants of a conversation;

identify the participants based on the one or more images;

localize relative positions of the participants based on the one or more images;

detect a plurality of speeches of the conversation through the microphone;

match one of the participants to each of the detected speeches according to the relative positions of the participants and count participations of the matched participant in the conversation;

identify a passive subject from all the participants according to the participations of all the participants in the conversation;

find a topic of the conversation between the participants; and

control the robot to move to align with the passive subject according to the relative position of the passive subject, and address the passive subject and speak a sentence related to the topic through the speaker to engage the passive subject;

wherein the conversation comprises a plurality of turns of dialogue; the counting the participations of the participant in the conversation comprises:

counting a total speaking time of the matched participant during a last αN turns of dialogue in the conversation, wherein α is an integer greater than 0, N is a total number of the participations; and

the conversation comprises a plurality of turns of dialogue; the identifying the passive subject from all the participants according to the participations of all the participants in the conversation comprises:

identifying the participant with the total speaking time below 2 standard deviations of an average speaking time in the conversation during the last αN turns of dialogue as the passive subject.

9. The electronic device of claim 8 , wherein the finding the topic of the conversation between the participants comprises:

transcribing each utterance in the speeches of the conversation into a text;

encoding the text corresponding to one or more spoken sentences in the conversation into a sentence-level semantic vector; and

selecting a fact with a closest similarity to the sentence-level semantic vector from a pre-selected fact collection, wherein the similarity meets a threshold; and

the speaking the sentence related to the topic through the speaker comprises:

speaking the sentence comprising the fact through the speaker.

10. The electronic device of claim 9 , wherein the fact is a trivia related to the topic, and the pre-selected fact collection is a corpus of topical facts comprising a plurality of single-sentence entries; and the selecting the fact with the closest similarity to the sentence-level semantic vector from the pre-selected fact collection comprises:

comparing a vector of each single-sentence entry in the corpus of topical facts with the sentence-level semantic vector; and

taking the single-sentence entry of the vector with a closest cosine similarity to the sentence-level semantic vector as the trivia; and

the speaking the sentence related to the topic through the speaker comprises:

speaking the sentence comprising the trivia through the speaker.

11. The electronic device of claim 9 , wherein the detecting the speeches of the conversation through the microphone comprises:

receiving an audio through the microphone; and

detecting the speeches of the conversation from the audio using voice activity detection to send a detected audio to an automatic speech recognition engine; and

the transcribing each utterance in the speeches of the conversation into the text comprises:

transcribing, through the automatic speech recognition engine, each utterance in the speeches of the conversation into the text based on the detected audio.

12. The electronic device of claim 9 , wherein the speaking the sentence comprising the fact through the speaker comprises:

formatting the sentence comprising the fact using natural language processing; and

converting the sentence into speech using a text-to-speech convertor to speak through the speaker.

13. The electronic device of claim 8 , wherein the counting the participations of the participant in the conversation further comprises:

counting a number of participation of the matched participant during the last αN turns of dialogue in the conversation; and

the identifying the passive subject from all the participants according to the participations of all the participants in the conversation further comprises:

identifying the participant with the number of participation less than β times during the last αN turns of dialogue as the passive subject, wherein β is an integer smaller than α and greater than 0.

14. A non-transitory computer readable storage medium storing one or more programs, wherein the one or more programs comprise instructions, which when executed by a robot having a camera, a microphone, and a speaker, cause the robot to:

obtain, through the camera, one or more images of a plurality of participants of a conversation;

identify the participants based on the one or more images;

localize relative positions of the participants based on the one or more images;

detect a plurality of speeches of the conversation through the microphone;

match one of the participants to each of the detected speeches according to the relative positions of the participants and count participations of the matched participant in the conversation;

identify a passive subject from all the participants according to the participations of all the participants in the conversation;

find a topic of the conversation between the participants; and

control the robot to move to align with the passive subject according to the relative position of the passive subject, and address the passive subject and speak a sentence related to the topic through the speaker to engage the passive subject;

wherein the conversation comprises a plurality of turns of dialogue; the counting the participations of the participant in the conversation comprises:

counting a total speaking time of the matched participant during a last αN turns of dialogue in the conversation, wherein a is an integer greater than 0, N is a total number of the participations; and

the conversation comprises a plurality of turns of dialogue; the identifying the passive subject from all the participants according to the participations of all the participants in the conversation comprises:

identifying the participant with the total speaking time below 2 standard deviations of an average speaking time in the conversation during the last αN turns of dialogue as the passive subject.

15. The storage medium of claim 14 , wherein the finding the topic of the conversation between the participants comprises:

transcribing each utterance in the speeches of the conversation into a text;

encoding the text corresponding to one or more spoken sentences in the conversation into a sentence-level semantic vector; and

selecting a fact with a closest similarity to the sentence-level semantic vector from a pre-selected fact collection, wherein the similarity meets a threshold; and

the speaking the sentence related to the topic through the speaker comprises:

speaking the sentence comprising the fact through the speaker.

16. The storage medium of claim 15 , wherein the fact is a trivia related to the topic, and the pre-selected fact collection is a corpus of topical facts comprising a plurality of single-sentence entries; and the selecting the fact with the closest similarity to the sentence-level semantic vector from the pre-selected fact collection comprises:

comparing a vector of each single-sentence entry in the corpus of topical facts with the sentence-level semantic vector; and

taking the single-sentence entry of the vector with a closest cosine similarity to the sentence-level semantic vector as the trivia; and

the speaking the sentence related to the topic through the speaker comprises:

speaking the sentence comprising the trivia through the speaker.

17. The storage medium of claim 15 , wherein the detecting the speeches of the conversation through the microphone comprises:

receiving an audio through the microphone; and

detecting the speeches of the conversation from the audio using voice activity detection to send a detected audio to an automatic speech recognition engine; and

the transcribing each utterance in the speeches of the conversation into the text comprises:

transcribing, through the automatic speech recognition engine, each utterance in the speeches of the conversation into the text based on the detected audio.

18. The storage medium of claim 14 , wherein the counting the participations of the participant in the conversation further comprises:

counting a number of participation of the matched participant during the last αN turns of dialogue in the conversation; and

the identifying the passive subject from all the participants according to the participations of all the participants in the conversation further comprises:

identifying the participant with the number of participation less than β times during the last αN turns of dialogue as the passive subject, wherein β is an integer smaller than α and greater than 0.