Hi!
Last week during Build 2023, there were several amazing announcements. I really liked Kevin Scott (CTO and Executive VP of AI) โThe era of the AI Copilot.โ
Kevin discusses the remarkable advancements in AI and its profound impact worldwide. The focus on AI model applications and the Azure platform is truly awe-inspiring. The possibilities are endless!
As I watched the video, my mind was buzzing with ideas for my own podcast, โNo Tiene Nombre.โ And guess what? I am going to experiment with the Podcast Copilot feature discussed in the presentation.
The main repo for this is available here:
https://github.com/microsoft/podcastcopilot
And hey, the full process is amazing (more info in the repo):
- Given the podcast audio file, the Whisper model performs speech-to-text to generate a transcript of the podcast.
- Given this transcript, the Dolly 2 model extracts the name of the guest on the podcast.
- Given the guest name, the Bing Search Grounding API retrieves a bio for the guest from the internet.
- Given the transcript and guestโs bio, the GPT-4 model generates a social media post promoting the podcast episode.
- Given the social media post, we use GPT-4 to create a relevant DALL-E prompt.
- Given that DALL-E prompt, the DALL-E model generates a corresponding image for the post.
- Finally, the user has an opportunity to review the content before posting, and if approved, a LinkedIn plugin will post the social media copy and image to LinkedIn.
๐ค Will it work in a Spanish Podcast? And next steps.
So, No Tiene Nombre is in Spanish. Will the Copilot be able to handle it?โ My guess is yes, now itโs time to test it.
Also, I donโt have access to GPT-4 for personal projects (yet), so this will be a nice exercise on how to make this work using GPT-3.5 and, removing the image generation step.
As Iโm working on this, Iโll be sharing my experience.
Happy coding!
Greetings
El Bruno
More posts in my blog ElBruno.com.
More info in https://beacons.ai/elbruno
Top comments (0)