Generating meeting notes
Methods and systems provide for real-time meeting notes within a communication session. In one embodiment, the system connects to a communication session with a plurality of participants; receives, in real time during the communication session, updates to a transcript of a conversation between the participants produced during the communication session, the transcript including timestamps for a plurality of utterances associated with speaking participants; receives a request from one of the participants to generate a meeting note which includes a subset of utterances from the transcript; determines a starting timestamp and an ending timestamp for utterances representing the boundaries of the meeting note; generates, in real time during the communication session, a meeting note where the utterances for the starting timestamp and ending timestamp are the starting utterance and ending utterance of the meeting note, respectively; and presents to at least the participant, within a user interface, the generated meeting note.
1. A method, comprising:
receiving a request to generate a meeting note corresponding to a first timestamp for a meeting with one or more participants, wherein the request is initiated by a participant of the one or more participants at the first timestamp;
determining, based on a transcript of the meeting, a context of the meeting at the first timestamp according to contexts of utterances prior to the first timestamp and contexts of utterances subsequent to the first timestamp, the determining being performed by an artificial intelligence (AI) model trained via self-supervised pre-training;
determining a starting timestamp and an ending timestamp representing boundaries of the meeting note that correspond to the context of the meeting at the first timestamp;
generating the meeting note according to the starting timestamp and the ending timestamp; and
presenting the generated meeting note to the one or more participants.
2. The method of claim 1 , wherein presenting the generated meeting note to the one or more participants occurs in real time during the meeting.
3. The method of claim 2 , wherein the generated meeting note is editable by the participants in real time during the meeting.
4. The method of claim 1 , wherein the generated meeting note is presented after the meeting has terminated within an after-meeting UI.
5. The method of claim 4 , wherein the generated meeting note is presented within the after-meeting UI concurrently with one or more additional meeting notes generated during the meeting.
6. The method of claim 1 , wherein presenting the generated meeting note comprises presenting at least a subset of a transcript of the meeting such that the utterances within the boundaries of the meeting note are highlighted within the transcript.
7. The method of claim 6 , wherein the highlighted utterances within the boundaries of the meeting note are modifiable by a participant to adjust one or more of the starting timestamp and ending timestamp.
8. The method of claim 1 , further comprising:
receiving a request from a participant to edit one or more of the starting timestamp and the ending timestamp; and
determining, based on the request, one or more of a new starting timestamp and a new ending timestamp.
9. The method of claim 8 , wherein the new starting timestamp and new ending timestamp are ingested as training data for training a model to determine starting timestamps and ending timestamps for one or more future meeting notes.
10. The method of claim 1 , wherein the AI model is trained to: extract one or more content-based features; and
generate a prediction of the starting timestamp and the ending timestamp based at least in part on the one or more content-based features.
11. The method of claim 10 , wherein the AI model extracts the content-based features using one or more of: a TextTiling training model, and a language model via self-supervised pre-training.
12. The method of claim 1 , wherein the AI model is trained to: extract one or more multimodal features; and
generate a prediction of the starting timestamp and the ending timestamp based at least in part on the one or more multimodal features.
13. A communication system comprising one or more processors configured to perform operations:
receiving a request to generate a meeting note corresponding to a first timestamp for a meeting with one or more participants, wherein the request is initiated by a participant of the one or more participants at the first timestamp;
determining, based on a transcript of the meeting, a context of the meeting at the first timestamp according to contexts of utterances prior to the first timestamp and contexts of utterances subsequent to the first timestamp, the determining being performed by an artificial intelligence (AI) model trained via self-supervised pre-training;
determining a starting timestamp and an ending timestamp representing boundaries of the meeting note that correspond to the context of the meeting at the first timestamp;
generating the meeting note according to the starting timestamp and the ending timestamp; and
presenting the generated meeting note to the one or more participants.
14. The communication system of claim 13 , wherein determining the starting timestamp and the ending timestamp for the meeting note comprises:
extracting one or more conversational features via a rule-based model; and
determining the starting timestamp and the ending timestamp based at least in part on the conversational features.
15. The communication system of claim 13 , wherein determining the starting timestamp and the ending timestamp for the meeting note comprises employing a pairwise meeting note boundary to generate a list of candidates via a rule-based model, each candidate comprising a boundary pair of a starting utterance and an ending utterance representing a boundary for the meeting note.
16. The communication system of claim 15 , wherein determining the starting timestamp and the ending timestamp for the meeting note further comprises employing the pairwise meeting note boundary to generate rankings for the each boundary pairs via a random forest regressor.
17. The communication system of claim 13 , wherein the request to generate a meeting note is initiated by a participant by interactively selecting a UI element for saving a note.
18. The communication system of claim 13 , wherein the one or more processors are further configured to perform the operations of:
detecting one or more automated note taking conditions,
wherein the request to generate a meeting note is automatically initiated based on the detecting of the one or more automated note taking conditions.
19. A non-transitory computer-readable medium containing instructions comprising:
instructions for receiving a request to generate a meeting note corresponding to a first timestamp for a meeting with one or more participants, wherein the request is initiated by a participant of the one or more participants at the first timestamp;
instructions for determining, based on a transcript of the meeting, a context of the meeting at the first timestamp according to contexts of utterances prior to the first timestamp and contexts of utterances subsequent to the first timestamp, the determining being performed by an artificial intelligence (AI) model trained via self-supervised pre-training;
instructions for determining a starting timestamp and an ending timestamp representing boundaries of the meeting note that correspond to the context of the meeting at the first timestamp;
instructions for generating the meeting note according to the starting timestamp and the ending timestamp; and
instructions for presenting the generated meeting note to the one or more participants.
20. The non-transitory computer readable medium of claim 19 , further containing instructions comprising:
instructions for receiving a request from a participant to edit one or more of the starting timestamp and the ending timestamp; and
instructions for determining, based on the request, one or more of a new starting timestamp and a new ending timestamp.