Population & Growth
8%
60%
Not everyone cooks with voice on. Many users browse recipes silently or follow written steps.
Activity Distribution (sessions / month)
· drag sliders; the mix auto-normalizes if you click normalize.
Session Characteristics
28 min
12
25%
Speech density tuning
12%
18%
32%
30%
Remainder of session is idle/cooking time (no audio tokens). These shares × session length → billable audio minutes.
Pricing (Bedrock Nova Sonic)
Ancillary per-session costs
Nova Sonic public pricing: speech-in $0.0034/1K tok, speech-out $0.0136/1K tok. Audio streams ~80–90 tok/s.
At defaults ⇒ — / input min, — / output min, blended ≈ — / conversation min.
Users · Month 12
—
Sessions · Month 12
—
Cost / active Sonic user / mo
—
Total 12-mo cost
—
| Month | Users | Sonic users | Sessions | Audio-in min | Audio-out min | Speech cost | Text cost | Ancillary | Total |
|---|
Model: for each month, users grow compounding. Sonic users = users × adoption%. Each Sonic user generates sessions/month per their activity bucket.
Session audio = session_len × (user-speech% + agent-speech%), split chatty vs typical. Tokens = minutes × 60 × tokens/sec. Text cost covers system prompt, tool schemas, and recipe context per session.
Defaults sourced from: AWS Bedrock Nova Sonic public pricing; audio token rate from AWS docs (~80–90 tok/s each direction); session length from ChefConnect activeContext (typical recipe 25–35 min). Tune freely.
Caveat: Bedrock bills tokens not minutes; the "per-minute" display is derived. Real traffic varies ±30% from these curves. Validate with a 2-week CloudWatch sample before committing to a forecast.
Defaults sourced from: AWS Bedrock Nova Sonic public pricing; audio token rate from AWS docs (~80–90 tok/s each direction); session length from ChefConnect activeContext (typical recipe 25–35 min). Tune freely.
Caveat: Bedrock bills tokens not minutes; the "per-minute" display is derived. Real traffic varies ±30% from these curves. Validate with a 2-week CloudWatch sample before committing to a forecast.