Video-based chapter generation for a communication session

1 . A method, comprising:

receiving a transcript and video content of a conversation between participants produced during a communication session, the transcript comprising timestamps for a plurality of utterances associated with speaking participants;

extracting high-resolution versions and low-resolution versions of frames from the video content;

identifying text in the low-resolution versions of the frames;

extracting one or more pieces of textual content from the high-resolution versions of the frames of the video content;

segmenting frames of the video content into a plurality of contiguous topic segments based on a coherence score for each topic segment, wherein the coherence score for each topic segment is determined based on a content vector scoring function and utterance embeddings;

determining a title for each topic segment from one or more of: the transcript, and the extracted textual content;

assigning a category label for each topic segment from a prespecified list of category labels; and

transmitting, to one or more client devices, the list of topic segments with determined title and assigned category label for each of the topic segments.

2 . The method of claim 1 , further comprising:

for each of the topic segments, further segmenting the utterances within that topic segment.

3 . The method of claim 1 , further comprising:

processing the video content to extract, via optical character recognition (OCR), one or more titles visible within the frames of the video content.

4 . The method of claim 1 , further comprising:

merging one or more neighboring segments according to one or more merging rules.

5 . The method of claim 4 , wherein the one or more merging rules comprise at least merging neighboring segments with a same title.

6 . The method of claim 4 , wherein the one or more merging rules comprise at least merging neighboring segments when one of the segments does not meet a threshold length.

7 . The method of claim 1 , further comprising:

determining a title for each topic segment based on the transcript.

8 . The method of claim 7 , wherein determining the title for each topic segment comprises:

extracting one or more top phrases from each topic segment;

determining a ranking of the top phrases for each topic segment; and

determining the title for each topic segment based on the top ranked phrase for the topic segment.

9 . The method of claim 8 , wherein extracting the one or more top phrases from each topic segment is performed using parts-of-speech (POS) tagging.

10 . The method of claim 8 , wherein determining the ranking of the top phrases for each topic segment is performed using inverse document similarity techniques where each top phrase is compared to the transcript as a whole.

11 . The method of claim 8 , wherein determining the ranking of the top phrases for each topic segment comprises:

determining a similarity score between each top phrase and the text of the utterances within the segment.

12 . The method of claim 1 , wherein assigning the category label for each topic segment is performed using named entity recognition (“NER”) techniques.

13 . The method of claim 1 , wherein assigning the category label for each topic segment is performed using topic model inferencing techniques.

14 . The method of claim 1 , wherein assigning the category label for each topic segment is performed using one or more post-processing rules.

15 . The method of claim 1 , wherein assigning the category label for each topic segment is performed using one or more trained artificial intelligence (AI) models, the training of the AI models being performed using unsupervised clustering techniques.

16 . The method of claim 1 , wherein the list of topic segments further comprises a starting and ending timestamp for each topic segment.

17 . A communication system, comprising:

one or more processors configured to:

receive a transcript and video content of a conversation between participants produced during a communication session, the transcript comprising timestamps for a plurality of utterances associated with speaking participants;

extract high-resolution versions and low-resolution versions of frames from the video content;

identify text in the low-resolution versions of the frames;

extract one or more pieces of textual content from the high-resolution versions of the frames of the video content;

segment frames of the video content into a plurality of contiguous topic segments based on a coherence score for each topic segment, wherein the coherence score for each topic segment is determined based on a content vector scoring function and utterance embeddings;

determine a title for each topic segment from one or more of: the transcript, and the extracted textual content;

assign a category label for each topic segment from a prespecified list of category labels; and

transmit, to one or more client devices, the list of topic segments with determined title and assigned category labels for each of the topic segments.

18 . The communication system of claim 17 , wherein the one or more processors is further configured to:

merge one or more neighboring segments according to one or more merging rules.

19 . The communication system of claim 18 , wherein the one or more merging rules comprise at least merging neighboring segments with a same title.

20 . A non-transitory computer-readable medium containing instructions that when executed by one or more processors, cause the processor to perform operations comprising:

receiving a transcript and video content of a conversation between participants produced during a communication session, the transcript comprising timestamps for a plurality of utterances associated with speaking participants;

extracting high-resolution versions and low-resolution versions of frames from the video content;

identifying text in the low-resolution versions of the frames;

extracting one or more pieces of textual content from the high-resolution frames of the video content;

segmenting frames of the video content into a plurality of contiguous topic segments based on a coherence score for each topic segment, wherein the coherence score for each topic segment is determined based on a content vector scoring function and utterance embeddings;

determining a title for each topic segment from one or more of: the transcript, and the extracted textual content;

assigning a category label for each topic segment from a prespecified list of category labels; and

transmitting, to one or more client devices, the list of topic segments with determined title and assigned category labels for each of the topic segments.