Podcast Insider
Podcasts have recently become super popular. There are great experts ready to share their thoughts, questions that get us all bothered are asked, they are accessible and … there are tons of them. Contexxt AI and web teams got together to eliminate some of the pain points of podcast listeners, and find the way to help make decisions based on experts’ opinion and most recent news and trends.
It’s a platform that analyses trustworthy podcasts, like Bloomberg Surveillance, Goldman Sachs, Barron’s Numbers by turning speech to text, highlighting so called sentiments in the text: positive attitude or prediction, negative, neutral; there is also your Watchlists: stocks and entities you want to keep an eye on – as soons as the company is mentioned with a certain sentiment, you can see that the color of the entity changes. That might give a user heads-up: maybe it’s high time to sell or buy, or keep the stocks.
There are things everyone is discussing now, these events influence economics, investments, employments. To quickly see what experts are saying about these phenomena, users can check a cloud of trending topics, click on the topic and see who mentioned it, in what context, and what’s been said about the issue recently.
We understand that experts are different and the level of credibility may vary. So our system can now recognize the voice of speakers, and users can click the expert’s name, check what the person does and what their main area of expertise is in accordance with their linkedin profile and official sources.
What’s under the hood?
To make the experience as smooth as possible, AI worked on the following things:
- After capturing streaming audio from podcasts, speech is converted to text. We had to take care of word timing. It allows the users to follow the text as they listen to the audio. It also allows us to define when a specific speaker is talking to recognize their voice and say what the name of the speaker is, the text is also punctuated properly
- Sentiment analysis, which allows us to understand the sentiment of the sector of the text, and change the color specific of a stock or company in user watchlist. After that a user can see the justification: a highlighted sector of the text, and make a grounded decision
- Text classification: we can classify what sector each block of the text relates to (sport-economy-technology-etc)
- Keywords extraction. We highlight the words that are significant in the text block (e.g. investment, oil crisis, war), which helps us to give a short summary of the text. The keywords that are mentioned frequently get to the trending topics cloud. That’s how the users can see short summaries from experts on a given topic
- Speaker classifications and recognition is based on the unique values of the speaker’s voice so that when we hear the voice next time the system can say who is speaking. We also filter the noise
- NER (Named-entity recognition) – defining the terms, names of the company, entities, geography, etc in the text block
- Stocks names extraction helps find the name of the company and the name of the stock it relates to (Apple – AAPL)