News Audio Intelligence.
Finally structured.
Your competitors get email alerts when a keyword fires. You get structured weekly narratives, sentiment trends, and geographic breakdowns — automatically, from hundreds of live audio feeds.
The medium your monitoring tool isn't watching
The 2024 election cycle made one thing clear: a significant bloc of voters — younger audiences, suburban households, rural markets — get their political news from audio first. Local NPR affiliates, independent political podcasts, and talk radio. This is where opinions form before they reach the evening news.
Traditional media intelligence watches TV clips and press mentions. It sends you a keyword alert when a name fires. It doesn't tell you:
- What narrative is forming on local audio before it reaches national press?
- How are left, center, and right-leaning podcasts framing the same story differently?
- Is the tone improving or declining week over week across 90+ markets?
- Which local markets are driving coverage — and which are silent?
How we compare
| Capability | Other Intelligence Tools | Catchwind |
|---|---|---|
| Weekly narrative summaries | ||
| Narrative drift scoring | ||
| Sentiment tracking | Partial | |
| National vs. local breakdown | ||
| Chapter arc / story arcs | ||
| Person & organization mention tracking | ||
| Keyword alerts | ||
| Audio monitoring |
Built for
Political Campaigns
Every week: who framed your candidate as a leader vs. a liability, which markets are running more opponent coverage, how many episodes flagged crisis signals, and a plain-English brief on the story threads driving your coverage — from local radio to national programs.
PR & Comms Agencies
Know when your client's narrative is shifting before they see the press clip. Framing direction, sentiment trends, and a drift score showing when coverage enters a new arc — structured weekly, not a dump of clips.
News Organizations
Benchmark coverage across markets. See what your competitors are covering and how the story is evolving outside your footprint.
Campaign workflow — four deliverables, every week
Built on existing schema fields — no custom setup required.
Is your candidate being covered as a leader, a target, or a liability? Scored per episode, aggregated weekly with a full framing-type distribution.
Airtime comparison vs. your opponent, broken down by market. See which local markets are giving them more coverage and whether their narrative is improving.
Episodes flagged for crisis-signal language before they amplify across markets. Weekly count with direct links to the exact moments and quotes.
2–6 story threads in plain English: what the coverage angle is, what's driving it, and whether it shifted from the prior week — with a drift score.
How it works
Monitor
Read what was said on local radio and podcasts without listening to the broadcast. 500+ feeds across 90+ markets, queued within minutes of publish.
Extract
Jump straight to the moment your candidate or client was named — across hundreds of stations, timestamped and searchable.
Analyze
A plain-English weekly brief on how a story is moving, who's rising or falling in coverage, and how the narrative differs by region — instead of raw alert spam.
The technology behind it
Every structured insight on Catchwind is produced by a pipeline of purpose-built models — no human editors, no manual tagging.
Speech-to-Text (ASR)
Audio is transcribed using Whisper large-v3, run locally on every episode. Each word is time-coded — enabling direct links to the exact moment a name or topic appears in the recording.
Named Entity Recognition
A fine-tuned BERT model identifies every person, organization, and location mentioned across each transcript. Mentions are resolved to canonical entities and stored with precise timestamps.
Sentiment Analysis
A separate classification model scores how each episode frames the subject — positive, negative, or neutral — at the episode level and aggregated across markets week over week.
Semantic Clustering
Story summaries are embedded using sentence transformers (BGE-large) and clustered weekly with UMAP + HDBSCAN. This is how Catchwind groups coverage into coherent story threads without any manual labeling.
LLM Narrative Generation
GPT-4o synthesizes each week's clustered stories into plain-English summaries, drift scores, and chapter breaks — the structured briefs delivered to analysts, campaigns, and PR teams.
On-Device Inference
ASR and NER run entirely on-device with no cloud dependency in the ingestion path — keeping audio data private and enabling batch processing across 500+ feeds simultaneously.
Coverage
500+
Audio Feeds
137+
Local News
24
National Programs
90+
Markets