Swiss Radio Ad-Spot AI
Budget: $250 – $750 USD
I am starting from the idea stage and need a practical roadmap—and ideally a working proof-of-concept—for an AI system that can monitor live Swiss radio streams and automatically log every commercial break. The goal is to surface four key pieces of information in real time:
• the advertiser’s brand name & company type
• exact start and end times (to the second)
• the language of the spot (German, French, Italian)
• the nature of surrounding programming—sports, music, news, or talk—so I can see ad placement context
Because everything will come from live streams rather than pre-recorded files, the solution has to ingest HLS/MP3/AAC streams, perform low-latency speech-to-text, and combine audio fingerprinting or text matching to recognise brand names even when the copy changes. Classifying the adjacent content matters just as much as detecting the ad itself, so a lightweight segment classifier that can tag short bursts of sports commentary versus music beds is essential.
Here is what I need from you:
• A clearly documented tech architecture (Python is preferred, but I’m open) that chains together stream capture, automatic speech recognition (e.g. Whisper, Google Speech-to-Text), brand/ad matcher, language detector and a context classifier.
• An initial model or ruleset for identifying at least the top Swiss national advertisers so I can see the concept working.
• A small dashboard or CSV/JSON feed that outputs advertiser name, start-time, end-time, spot length, language, and surrounding-content tag in real time.
• Guidance on how to scale the approach from a single test station to dozens of Swiss stations running 24/7.
If you can deliver a lightweight but functional prototype that satisfies those bullet-point acceptance criteria, we can move on to a fuller production build.
• the advertiser’s brand name & company type
• exact start and end times (to the second)
• the language of the spot (German, French, Italian)
• the nature of surrounding programming—sports, music, news, or talk—so I can see ad placement context
Because everything will come from live streams rather than pre-recorded files, the solution has to ingest HLS/MP3/AAC streams, perform low-latency speech-to-text, and combine audio fingerprinting or text matching to recognise brand names even when the copy changes. Classifying the adjacent content matters just as much as detecting the ad itself, so a lightweight segment classifier that can tag short bursts of sports commentary versus music beds is essential.
Here is what I need from you:
• A clearly documented tech architecture (Python is preferred, but I’m open) that chains together stream capture, automatic speech recognition (e.g. Whisper, Google Speech-to-Text), brand/ad matcher, language detector and a context classifier.
• An initial model or ruleset for identifying at least the top Swiss national advertisers so I can see the concept working.
• A small dashboard or CSV/JSON feed that outputs advertiser name, start-time, end-time, spot length, language, and surrounding-content tag in real time.
• Guidance on how to scale the approach from a single test station to dozens of Swiss stations running 24/7.
If you can deliver a lightweight but functional prototype that satisfies those bullet-point acceptance criteria, we can move on to a fuller production build.