STT+RAG+LLM+TTS
Cost Simulator — Voice Pipeline
Estimate the real monthly cost of your voice pipeline (STT + RAG + LLM + TTS) accounting for subscription plans, included minutes, top-ups, and parallel stream limits.
Start with a preset
Usage Volume
Set the number of users and session duration — total minutes are computed automatically. Parallel streams define simultaneous capacity and may trigger plan alerts.
Users / month250
1050010k
Avg session duration10 min
2 min30 min120 min
Simultaneous parallel streams10 streams
1 stream50 streams1000 streams
Users/month
250
Avg session
10 min
Total minutes/month
2.5k
250 × 10 min
Parallel streams
10
simultaneous capacity
Formula: Monthly variable cost = (cost/min STT + RAG + LLM + TTS) × 250 users × 10 min/session = 2,500 total min/month. Parallel streams (10) define simultaneous capacity and may require a specific subscription plan.
1 — Speech Recognition (STT)
2 — Knowledge Base (RAG)
Monthly fixed cost: $25/mo (vector DB) — independent of volume
3 — Language Model (LLM)
Assumption: 2 turns/min, ~2,000 input tokens + ~150 output tokens per turn.
4 — Voice Synthesis (TTS)
Simulation Results
Total cost / month (variable)
$455
250 users × 10 min + $25 RAG
Optimal cost / month (with plans)
$455
Best STT + TTS + LLM plan
Cost / user
$1.72
10 min × $0.172/min
Estimated pipeline latency
625ms
STT + RAG + LLM + TTS
Recommended optimal plan for your volume(2,500 min/mois)
STT — Deepgram Nova-3
Pay As You Go
overage $0.00430/min
$11/mo
LLM — Gemini 2.0 Flash
Pay As You Go
overage $0.00052/min
$1/mo
The “Free (Google AI Studio)” plan includes 0 min but allows no overage. Your volume (2,500 min/mo) exceeds this by 2,500 min — “Pay As You Go” plan selected automatically.
TTS — Eleven v4 / v4 Turbo
Scale
$299/mo + overage $0.170/min
$418/mo
The “Free” plan would cost $896.40/mo with overage. The “Scale” plan ($299/mo + 1,800 min included) costs $418.00/mo — saving $478.40/mo.
Variable cost breakdown
TTS 97%
| Component | Provider | $/min (optimal plan) | Selected plan | $/user (plan) | Total/mo |
|---|---|---|---|---|---|
| STT | Deepgram Nova-3 | $0.00430 | Pay As You Go | $0.043 | $11 |
| RAG | Voyage AI + Supabase pgvector | $0.00000 | Free + Supabase Free | $0.00000 | $0.00+$25 |
| LLM | Gemini 2.0 Flash | $0.00052 | Pay As You Go | $0.00520 | $1 |
| TTS | Eleven v4 / v4 Turbo | $0.167$0.200 | Scale$299/mo | $1.67 | $418 |
| Total (optimal plans) | $1.72 | $455 | |||
| OPTIMAL TOTAL (best plans) | — | $455 | |||
Pipeline latency breakdown (estimated)
75ms
STT
50ms
RAG
300ms
LLM TTFT
200ms
TTS TTFA
Estimated total:625ms⚠ Acceptable but optimizable
Preset comparison at your volume
Estimated monthly cost for 250 users × 10 min/session = 2,500 min/month. 10 parallel streams.
| Preset | STT | RAG | LLM | TTS | $/user | Total/mo | Latency | EU | Streams |
|---|---|---|---|---|---|---|---|---|---|
Storygami — Optimal | $11 | $0.00+$25 | $1 | $500 | $2.05 | $537 | 625ms | ∞ | |
Storygami — Budget | $4 | $0.00+$25 | $1 | $31 | $0.147 | $62 | 542ms | ∞ | |
Edugami — Optimal | $5 | $0.00+$25 | $1 | $50 | $0.225 | $81 | 769ms | ∞ | |
Sovereign EU Stack | $26 | $0.00+$25 | $1 | $108 | $0.542 | $160 | 686ms | 60 | |
Self-Hosted (Open-Source) | $0.00 | $0.00 | $0.00 | $2 | $0.00700 | $2 | 650ms | ∞ |
Full STT comparison
15 engines, WER benchmarks
Full TTS comparison
25 engines, ELO, expressiveness
Interactive V2V Pipeline
Latency & component diagram