# Methodology — texttospeech.com

> How the meta-leaderboard is computed. We aggregate, we don't test. Every number traces to a public source.

## Key Information
- Methodology version: v2.0
- Boards aggregated: 3
- Scoring method: Bayesian Pairwise Rank Aggregation (BPRA)

## The Boards
Currently aggregating: artificialanalysis.com (Speech Arena Elo), voicearena.com (arena Elo, US English), humannessindex.vapi.ai (humanness score 0–100)

## Scoring Process
- Fractional scaling: Rank #1 = 1.0, last row = 1/N (larger boards worth more)
- Bayesian coverage smoothing: Models on few boards pulled toward field average
- Ecosystem-normalized Borda count: Competitors defeated across all boards
- Schulze tiebreak: Strongest chain of board wins decides ties

## Confidence Labels
- High: On ≥2/3 boards with broad agreement
- Med: On 2+ boards
- Low: Single board or disagreeing boards

## Independence
No vendor sponsorship, referral fees, or paid placement. Every engine goes through the identical formula.

## Links
- Full methodology: https://texttospeech.com/methodology
- Methodology whitepaper: https://texttospeech.com/blog/our-methodology
- Leaderboard: https://texttospeech.com/