Live source snapshot
MAI-Transcribe-2-Streaming on AA-WER
Artificial Analysis reports 2.5% word error rate for final and first-partial transcripts from MAI-Transcribe-2-Streaming.
Source snapshots refreshed
Five measurement lanesVOICE / S2S
Reasoning, task completion, conversation dynamics, experience, and latency stay separate.
Benchmark profile
- Scope
- One exact Microsoft AI hosted configuration captured October 1, 2026; AA-AgentTalk, VoxPopuli, and Earnings22
- Primary metric
- Word error rate in percent and latency after speech end in seconds; lower is better
- Owner
- Artificial Analysis
- Available evidence
- Exact evaluator payload retained in private evidence; percentages rounded to one decimal and seconds to two decimals
What it measures
Voice experience
Is the exchange natural, robust, and responsive?
Latency
How long do responses, tool calls, and full tasks take?
Available results
- Final transcript WER
- 2.5%
- Time to final after speech end
- 0.13 s
- First-partial WER after speech end
- 2.5%
- Time to first partial after speech end
- 0.12 s
Evaluator measurement
Payload ratio 0.025136; lower is better.
Evaluator measurement
Payload 0.1275691718681629 seconds.
Evaluator measurement
Payload ratio 0.02516275142251098.
Evaluator measurement
Payload 0.12193524507770948 seconds.
Interpretation limit
This single configuration capture does not mirror the full leaderboard. The index weights AA-AgentTalk at 50%, VoxPopuli at 25%, and Earnings22 at 25%. Latency starts at SileroVAD-detected speech end; first partial is the first transcript-bearing event afterward. Missing endpoint finals can use the documented partial fallback. Evaluation date, hardware, and confidence intervals are not published in this row. The capture date does not establish when the evaluation ran.