# Voice Call Noise-Suppression Telemetry Dashboard

## PURPOSE
A Streamlit dashboard that visualizes a single voice-agent call and shows exactly where
background speaker suppression engaged. The Dylos demo plays live background speech while
the agent responds, so the honest question is: *which moments in this call were caller
turns and which were TV, radio, or announcement audio — and did suppression catch them?*
This dashboard turns a call log into an interactive timeline so anyone can see it.

## WHY THIS MATTERS
The video sells a feature you can only judge by ear. A dashboard makes it judgeable by
eye and by number: you can point at the exact second the radio ad started and see whether
the suppression gate fired, and how much energy leaked through. It is the evidence layer
that lets a team trust — or reject — a suppression setting before shipping it.

## HOW TO BUILD
Use Streamlit for the app and Plotly for the charts. Steps: (1) load a per-frame CSV of
call telemetry (`t`, `caller_energy`, `background_energy`, `suppressed`);
(2) render a stacked time-series of caller vs. background energy; (3) overlay suppression
windows as shaded spans using `add_vrect`; (4) add sidebar filters for call id and a
leakage threshold. Key snippet:

```python
fig = go.Figure()
fig.add_trace(go.Scatter(x=df.t, y=df.caller_energy, name="caller"))
fig.add_trace(go.Scatter(x=df.t, y=df.background_energy, name="background"))
for start, end in suppression_spans(df):
    fig.add_vrect(x0=start, x1=end, fillcolor="green", opacity=0.15, line_width=0)
st.plotly_chart(fig, use_container_width=True)   # TODO: add per-call KPI cards
```

## WHEN TO USE IN PRODUCTION
Great for support engineers diagnosing "the bot answered my TV" tickets, for QA reviewing
flagged calls, and for demos to stakeholders before a suppression rollout. Use it when a
human needs to inspect a handful of calls, not thousands. Do **not** use it as a batch
reporting system — pick the CLI analyzer for that. Avoid it for real-time monitoring;
Streamlit is a request/response UI, not a streaming console.

## HOW IT HELPS STUDENTS
Builds intuition for audio telemetry, Plotly time-series overlays, Streamlit layout, and
the product-thinking skill of turning model output into something a human can audit.

## STARTER CODE
See `starter/main.py` for a runnable Streamlit app with a deterministic synthetic call.
