Skip to main content

MicrosoftAudio modelPreview

MAI-Voice-2-Flash

Popularityno data yet

n/ano search or usage data

Summary

MAI-Voice-2-Flash is the faster and less expensive speech synthesis variant of MAI-Voice-2. Microsoft describes roughly twice the speed and a price of $15 per 1M characters in the Flash announcement.

The model identifier is mai-voice-2-flash, with Foundry access in preview. Voice 2 Flash is the earlier published Flash variant, separate from the later MAI-Voice-2.1-Flash generation.

GraduallyBench Audio

In the speech output text comparison, the re-transcription of MAI-Voice-2-Flash differs from the source by 3.7 % (12 errors over 321 words, German and English). The median of the 20 models with a German and English value is 3.5 %.

Word error rate, German and English

3.7 %

12 errors over 321 reference words

Character error rate

3.1 %

53 errors over 1,685 characters

Scope

16 / 16

valid of planned outputs

Access

OpenRouter

Text comparison, German and English, compared

Differing words after re-transcription in percent (German and English), lower is better

20 of 20 models

Smaller test scope: Fish Audio S2 Pro, Fish Audio S2.1 Pro, MAI-Voice-2, MAI-Voice-2-Flash, Sonic 3.6, Voxtral Mini TTS, GPT-4o Mini TTS.

As of 11/10/2026. Source: GraduallyBench Audio

MAI-Voice-2-Flash

Benchmark run
  • The text comparison uses a second speech recognizer as a proxy.
  • Naturalness and voice quality were not evaluated.
  • The voice actually used is not confirmed.
  • The route does not confirm the exact model version.
Show test details

Speech output with text comparison

OpenRouter · German, English

16 valid outputs from 16 attempted and 16 planned test slots.

Speech output text comparison, WER
4.35 %(7 / 161 reference words)German
Speech output text comparison, CER
3.7 %(33 / 891 reference characters)German
Speech output text comparison, WER
3.13 %(5 / 160 reference words)English
Speech output text comparison, CER
2.52 %(20 / 794 reference characters)English
Speech output text comparison, WER
3.74 %(12 / 321 reference words)German, English
Speech output text comparison, CER
3.15 %(53 / 1,685 reference characters)German, English

WER compares words and CER compares characters with the reference. The speech output was re-transcribed with GPT Transcribe. Its exact backend revision is unconfirmed. Read speech and digital silence do not cover spontaneous speech or real background noise. Error rates can exceed 100 %. Where used, the test references were not independently reviewed by humans. Test coverage and methods differ by model. Each metric stands on its own; there is no combined overall score across tests. Data compiled on 11/10/2026. How we test · Download all Decision and Audio results as JSON

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 3 published benchmark results

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

Design Arena Audio Realism (Show measurement, test conditions, and source)
Source value
1,126
Metric
Bradley-Terry Elo rating
Unit
elo
Category
audio-quality
Direction
Higher is better
Sample
151
Source type
official-leaderboard
Evaluator
Design Arena Audio Realism leaderboard
Status
superseded
Retrieved at
2026-10-17
Methodology
Official benchmark leaderboard. Captured public leaderboard API request: POST /api/leaderboard with {"arenaType":"models","category":"audiorealism","variationName":"public"}. The exact mai-voice-2-flash row reports wins=74, losses=77, battles=151, winRate=49, elo=1126, btStdErr=null. Audio-realism judgments use blinded A/B comparison for naturalness. The page describes one female and one male conversational American-English voice per model; it does not identify exact voice IDs or an immutable served revision.
Context
No confidence interval is published (btStdErr is null); exact voice IDs and immutable served revision are also absent. No speed, price, or 2.1-Flash data is used.
Source Design Arena Audio Realism leaderboard
DesignArena Audio Realism Elo (Show measurement, test conditions, and source)
Source value
1,137
Metric
Bradley-Terry preference Elo for perceived human-like voice realism
Unit
elo
Benchmark version
DesignArena Audio Realism public board snapshot 2026-10-28
Direction
Higher is better
Source type
independent-evaluator
Evaluator
DesignArena
Status
active
Retrieved at
2026-10-28
Methodology
Independent evaluation. Current public Audio Realism row mai-voice-2-flash; one male and one female American-English voice selected per model; blind same-prompt, same-gender comparisons using a fixed held-out set of 500+ versioned prompts across phone-agent, explainer and conversation categories; 166 battles.
Source DesignArena
DesignArena TTS Arena Elo (Show measurement, test conditions, and source)
Source value
1,118
Metric
blind pairwise preference Elo
Unit
elo
Benchmark version
DesignArena public TTS board snapshot 2026-10-28
Direction
Higher is better
Source type
independent-evaluator
Evaluator
DesignArena
Status
active
Retrieved at
2026-10-28
Methodology
Independent evaluation. Current DesignArena public TTS row mai-voice-2-flash; same-prompt pairwise human preference; 542 battles. Board update 2026-10-28 16:06 UTC.
Context
Battles are pairwise-comparison appearances, not independent prompt count; no row-level confidence interval published.
Source DesignArena

All specifications

Architecture

Model class
Low-latency speech synthesis
Source: Model class

Input and output

Input
Text and reference audio
Source: Input
Output
Audio
Source: Output
Speed
About twice as fast as MAI-Voice-2
Source: Speed

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking
Unknown
no source
Watermarking technique
Not known
no source

Licence and access

Access
Microsoft Foundry API
Source: Access
Shutdown
None announced
no source

API

API model ID
mai-voice-2-flash
Source: API model ID

Further details

Voice cloning
Reference audio with consent guardrails
Source: Voice cloning

Prices

Price
$15 per 1M charactersSource

Sources and data date

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.

Without vendor claims and historical or disputed results; those stay labelled in the single values.

TypeEvidence and data date
Research statusResearch date October 4, 2026. 9 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources

quality-benchmarks

Original wording

Voice-2 preference results and Voice-2.1-Flash latency cannot be transferred to this exact earlier Flash checkpoint.

Unresolved

quality-benchmarks

Note

Design Arena's Audio Realism leaderboard reports a Bradley-Terry Elo preference rating of 1,126 for a row named MAI Voice 2 Flash. The sample uncertainty is not published (btStdErr is null), and the evaluated conversational voices have no public exact IDs or immutable served revision. This preference result is not transferred to MAI Voice 2 or MAI Voice 2.1 Flash.

Not found in the inspected sources

quality-benchmarks

Note

No numeric quality score or exact Artificial Analysis row was found for older MAI-Voice-2-Flash. Microsoft's relative '2x faster' statement has no published absolute latency setup here and is not a quality benchmark; MAI-Voice-2.1-Flash is a distinct successor.

Not published

quality-benchmarks

Note

Microsoft's exact MAI-Voice-2-Flash launch announcement provides relative speed and price claims but no numeric quality score or evaluation denominator. Current MAI-Voice-2 human-preference results and the updated 2.1-Flash page concern different variants and were not transferred.

Not published

confidence-interval

Original wording

The exact leaderboard row has btStdErr=null and publishes no confidence interval.

Unresolved

served-voice-identity

Original wording

The benchmark describes two conversational American-English voices per model but publishes neither exact MAI Voice 2 Flash voice IDs nor an immutable served revision.

Additional sourceMicrosoft MAI Voice 2 Flash announcement (retrieved October 3, 2026; October 4, 2026; October 15, 2026; October 16, 2026) · Microsoft MAI Voice 2 Flash announcement · Editorial description reviewed October 3, 2026
Additional sourceMicrosoft MAI Voice 2 announcement (retrieved October 3, 2026; October 4, 2026) · Microsoft MAI Voice 2 announcement · Editorial description reviewed October 3, 2026
Additional sourceDesign Arena Audio Realism leaderboard (retrieved October 17, 2026)
Additional sourceDesignArena (retrieved October 28, 2026)12
Additional sourceModel metadata (retrieved October 17, 2026; October 28, 2026)
Additional sourceModel metadata (retrieved October 17, 2026)