Method and system for generating podcast metadata to facilitate searching and recommendation

1. A method comprising:

obtaining, by a computing system, a text representation of a podcast episode;

obtaining, by the computing system, person data defining a list of person names;

correlating, by the computing system, the person data with the text representation of the podcast episode, to find a match between (i) a person name from the list of the person names and (ii) a text string in the text representation of the podcast episode;

predicting, by the computing system, a named-entity span in the text representation of the podcast episode;

determining, by the computing system, that the predicted named-entity span matches a location of the text string in the text representation of the podcast episode;

based on the determining that the predicted named-entity span matches the location of the text string in the text representation of the podcast episode, generating, by the computing system, metadata that associates the person name with the podcast episode; and

outputting, by the computing system, the generated metadata.

2. The method of claim 1 , wherein the person names in the list defined by the person data are names of people, the method further comprising:

filtering the person data based on accreditations of the people.

3. The method of claim 1 , wherein the person name is a name of a person, the method further comprising:

using, the computing system, machine-learning-based role identification to determine a role of the person in the podcast episode;

based on the determined role of the person in the podcast episode, generating, by the computing system, additional metadata that identifies the role of the person in the podcast episode; and

outputting, by the computing system, the generated additional metadata.

4. The method of claim 3 , wherein the role comprises at least one item selected from the group consisting of a host of the podcast episode, a guest of the podcast episode, and a subject of the podcast episode.

5. The method of claim 1 , wherein the person name is a name of a person, the method further comprising:

making a voice-identification-based determination of whether a voice of the person is included in the podcast episode, wherein the generating of the metadata that associates the person name with the podcast episode is further based on the voice-identification-based determination.

6. The method of claim 1 , wherein the person name is a name of a person, the method further comprising:

using, by the computing system, voice identification as a basis to determine one or more times in the podcast episode when a voice of the person is present; and

generating and outputting, by the computing system, additional metadata indicating the one or more determined times.

7. The method of claim 1 , further comprising using the generated metadata as a basis to facilitate podcast searching based on the person name.

8. A computing system comprising:

one or more processors;

non-transitory data storage; and

program instructions stored in the non-transitory data storage and executable by the one or more processors to carry out operations including:

obtaining a text representation of a podcast episode,

obtaining person data defining a list of person names,

correlating the person data with the text representation of the podcast episode, to find a match between (i) a person name from the list of the person names and (ii) a text string in the text representation of the podcast episode,

predicting a named-entity span in the text representation of the podcast episode,

determining that the predicted named-entity span matches a location of the text string in the text representation of the podcast episode,

based on the determining that the predicted named-entity span matches the location of the text string in the text representation of the podcast episode, generating metadata that associates the person name with the podcast episode, and

outputting the generated metadata.

9. The computing system of claim 8 , wherein the person names in the list defined by the person data are names of people, the operations further including:

filtering the person data based on accreditations of the people.

10. The computing system of claim 8 , wherein the person name is a name of a person, the operations further including:

using machine-learning-based role identification to determine a role of the person in the podcast episode;

based on the determined role of the person in the podcast episode, generating additional metadata that identifies the role of the person in the podcast episode; and

outputting the generated additional metadata.

11. The computing system of claim 10 , wherein the role comprises at least one item selected from the group consisting of a host of the podcast episode, a guest of the podcast episode, and a subject of the podcast episode.

12. The computing system of claim 8 , wherein the person name is a name of a person, the operations further including:

making a voice-identification-based determination of whether a voice of the person is included in the podcast episode, wherein the generating of the metadata that associates the person name with the podcast episode is further based on the voice-identification-based determination.

13. The computing system of claim 8 , wherein the person name is a name of a person, the operations further including:

using voice identification as a basis to determine one or more times in the podcast episode when a voice of the person is present; and

generating and outputting additional metadata indicating the one or more determined times.

14. The computing system of claim 8 , further comprising using the generated metadata as a basis to facilitate podcast searching based on the person name.

15. A non-transitory computer-readable medium having stored thereon program instructions that, upon execution by one or more processors, cause performance of a set of operations comprising:

obtaining a text representation of a podcast episode;

obtaining person data defining a list of person names;

correlating the person data with the text representation of the podcast episode, to find a match between (i) a person name from the list of the person names and (ii) a text string in the text representation of the podcast episode;

predicting a named-entity span in the text representation of the podcast episode;

determining that the predicted named-entity span matches a location of the text string in the text representation of the podcast episode;

based on the determining that the predicted named-entity span matches the location of the text string in the text representation of the podcast episode, generating metadata that associates the person name with the podcast episode; and

outputting the generated metadata.

16. The non-transitory computer-readable medium of claim 15 , wherein the person names in the list defined by the person data are names of people, the operations further comprising:

filtering the person data based on accreditations of the people.

17. The non-transitory computer-readable medium of claim 15 , wherein the person name is a name of a person, the operations further comprising:

using machine-learning-based role identification to determine a role of the person in the podcast episode;

based on the determined role of the person in the podcast episode, generating additional metadata that identifies the role of the person in the podcast episode; and

outputting the generated additional metadata.

18. The non-transitory computer-readable medium of claim 17 , wherein the role comprises at least one item selected from the group consisting of a host of the podcast episode, a guest of the podcast episode, and a subject of the podcast episode.

19. The non-transitory computer-readable medium of claim 15 , wherein the person name is a name of a person, the operations further including:

making a voice-identification-based determination of whether a voice of the person is included in the podcast episode, wherein the generating of the metadata that associates the person name with the podcast episode is further based on the voice-identification-based determination.

20. The non-transitory computer-readable medium of claim 15 , wherein the person name is a name of a person, the operations further including:

using voice identification as a basis to determine one or more times in the podcast episode when a voice of the person is present; and

generating and outputting additional metadata indicating the one or more determined times.