MicrosoftAudio modelPreview
MAI-Voice-2.1
PopularityRank 649 of 875
1out of 100-1 vs. last month
Summary
MAI-Voice-2.1 turns text into natural-sounding speech across 23 languages and 26 locales. A short reference recording can guide the voice, with Microsoft documenting consent guardrails for cloning.
Access is provided through Microsoft Foundry as a public preview without an SLA. Microsoft lists a price of $22 per 1M characters. MAI-Voice-2.1 is for generated audio, not transcription or general audio understanding.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 9 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.
Cekura TTS hard-part accuracy: numbers and shorthand (Show measurement, test conditions, and source)
- Source value
- 85
- Score
- 85
- Metric
- Share of hard-part readings accepted for numbers and shorthand
- Unit
- %
- Benchmark version
- Cekura TTS Model Benchmarks; current page snapshot 2026-10-19; run date and serving revision not stated
- Category
- audio
- Direction
- Higher is better
- Source type
- official-leaderboard
- Evaluator
- Cekura Bench TTS Model Benchmarks
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Official benchmark leaderboard. Evaluator-run TTS benchmark. Cekura says each model read the same 104 agent sentences three times from one US East cloud region, using one stock voice and its own streaming protocol. Non-streaming timing runs from sending the full text to first audible sound; streaming timing runs from the last word sent to first audible sound. Reading accuracy is estimated from Deepgram Nova-3 and OpenAI Whisper transcripts, counting accepted readings without human review. Naturalness uses UTMOS, an open model predicting a listening-test score. The captured rows do not state per-cell N, CI, exact run date, serving/API revision, or exact voice ID unless shown separately.
- Context
- Cekura discloses that it runs the benchmark itself and sells testing and monitoring tools in the same market. The results are evaluator-published and include voice, runtime, protocol, ASR, and UTMOS effects; they are not a human-only or isolated-model evaluation.
Cekura TTS hard-part accuracy: codes and addresses (Show measurement, test conditions, and source)
- Source value
- 67
- Score
- 67
- Metric
- Share of hard-part readings accepted for codes and addresses
- Unit
- %
- Benchmark version
- Cekura TTS Model Benchmarks; current page snapshot 2026-10-19; run date and serving revision not stated
- Category
- audio
- Direction
- Higher is better
- Source type
- official-leaderboard
- Evaluator
- Cekura Bench TTS Model Benchmarks
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Official benchmark leaderboard. Evaluator-run TTS benchmark. Cekura says each model read the same 104 agent sentences three times from one US East cloud region, using one stock voice and its own streaming protocol. Non-streaming timing runs from sending the full text to first audible sound; streaming timing runs from the last word sent to first audible sound. Reading accuracy is estimated from Deepgram Nova-3 and OpenAI Whisper transcripts, counting accepted readings without human review. Naturalness uses UTMOS, an open model predicting a listening-test score. The captured rows do not state per-cell N, CI, exact run date, serving/API revision, or exact voice ID unless shown separately.
- Context
- Cekura discloses that it runs the benchmark itself and sells testing and monitoring tools in the same market. The results are evaluator-published and include voice, runtime, protocol, ASR, and UTMOS effects; they are not a human-only or isolated-model evaluation.
Cekura TTS hard-part accuracy: read from context (Show measurement, test conditions, and source)
- Source value
- 73
- Score
- 73
- Metric
- Share of hard-part readings accepted for context-dependent readings
- Unit
- %
- Benchmark version
- Cekura TTS Model Benchmarks; current page snapshot 2026-10-19; run date and serving revision not stated
- Category
- audio
- Direction
- Higher is better
- Source type
- official-leaderboard
- Evaluator
- Cekura Bench TTS Model Benchmarks
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Official benchmark leaderboard. Evaluator-run TTS benchmark. Cekura says each model read the same 104 agent sentences three times from one US East cloud region, using one stock voice and its own streaming protocol. Non-streaming timing runs from sending the full text to first audible sound; streaming timing runs from the last word sent to first audible sound. Reading accuracy is estimated from Deepgram Nova-3 and OpenAI Whisper transcripts, counting accepted readings without human review. Naturalness uses UTMOS, an open model predicting a listening-test score. The captured rows do not state per-cell N, CI, exact run date, serving/API revision, or exact voice ID unless shown separately.
- Context
- Cekura discloses that it runs the benchmark itself and sells testing and monitoring tools in the same market. The results are evaluator-published and include voice, runtime, protocol, ASR, and UTMOS effects; they are not a human-only or isolated-model evaluation.
Cekura TTS hard-part accuracy: medical terms (Show measurement, test conditions, and source)
- Source value
- 99
- Score
- 99
- Metric
- Share of hard-part readings accepted for medical terms
- Unit
- %
- Benchmark version
- Cekura TTS Model Benchmarks; current page snapshot 2026-10-19; run date and serving revision not stated
- Category
- audio
- Direction
- Higher is better
- Source type
- official-leaderboard
- Evaluator
- Cekura Bench TTS Model Benchmarks
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Official benchmark leaderboard. Evaluator-run TTS benchmark. Cekura says each model read the same 104 agent sentences three times from one US East cloud region, using one stock voice and its own streaming protocol. Non-streaming timing runs from sending the full text to first audible sound; streaming timing runs from the last word sent to first audible sound. Reading accuracy is estimated from Deepgram Nova-3 and OpenAI Whisper transcripts, counting accepted readings without human review. Naturalness uses UTMOS, an open model predicting a listening-test score. The captured rows do not state per-cell N, CI, exact run date, serving/API revision, or exact voice ID unless shown separately.
- Context
- Cekura discloses that it runs the benchmark itself and sells testing and monitoring tools in the same market. The results are evaluator-published and include voice, runtime, protocol, ASR, and UTMOS effects; they are not a human-only or isolated-model evaluation.
Cekura MAI-Voice-2.1 estimated hard-parts reading percentage (Show measurement, test conditions, and source)
- Source value
- 82.2
- Score
- 82.2
- Metric
- Cekura task-labelled hard-parts reading percentage estimated from speech-to-text transcripts; scoring formula and denominator not disclosed
- Unit
- %
- Benchmark version
- Cekura TTS benchmark page, last updated 2026-10-19; exact run date not stated
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Cekura Bench
- Status
- active
- Retrieved at
- 2026-10-24
- Methodology
- Independent evaluation. Cekura names MAI-Voice-2.1 and displays 19 fixed reading sentences for that voice. Its 82.2% figure is estimated from speech-to-text transcripts, not a human listening test. Cekura does not disclose the scoring formula, denominator, ASR engine(s), language, exact request model ID/API route, serving revision, or confidence interval. The page was last updated 2026-10-19 and credits methodology to Dileep Chagam.
- Context
- This is Cekura's task-labelled reading percentage, not a general audio-quality or latency score. The page shows 19 sentences, but that does not establish the scoring denominator or metric N; 26 is the voice-comparison cohort, not N. Formula, denominator, ASR scorer identity, language, exact API route/serving revision, and confidence interval are not disclosed.
Microsoft-published model-inference latency (Show measurement, test conditions, and source)
- Source value
- 0.55
- Score
- 0.55
- Metric
- approximate model-inference latency
- Unit
- seconds
- Benchmark version
- Microsoft MAI-Voice-2.1 model page retrieved 2026-10-16
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Microsoft AI
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The model page labels this as model-inference latency and reports approximately 550 ms. The page does not specify prompt length, output length, hardware, request count, or measurement method.
- Context
- This is an approximate vendor-published latency figure, not a standardized quality measurement or independently reproduced benchmark.
Cekura TTS naturalness score (Show measurement, test conditions, and source)
- Source value
- 4.18
- Score
- 4.18
- Metric
- Naturalness rating reported on Cekura's text-to-speech benchmark page
- Unit
- rating
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Cekura Bench TTS Model Benchmarks
- Status
- active
- Retrieved at
- 2026-10-21
- Methodology
- Independent evaluation. Cekura displays the first unedited output for 19 test sentences. The page does not disclose the scoring method, serving revision, run date, or evaluated sample count.
- Context
- The page reports a model-specific naturalness rating but does not state its scale or scoring method, whether listeners or a predictor produced it, an evaluated N, confidence interval, run date, or exact serving revision.
MAI-Voice-2.1 median speech-start latency (Show measurement, test conditions, and source)
- Source value
- 861
- Score
- 861
- Metric
- Time to start speaking on a typical sentence, median
- Unit
- ms
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Cekura Bench
- Status
- active
- Retrieved at
- 2026-10-23
- Methodology
- Independent evaluation. Cekura reports MAI-Voice-2.1 speech-start latency for a typical sentence and the slowest one in ten. The page does not state a measurement date, utterance count, or immutable serving revision.
- Context
- Cekura says it runs these benchmarks itself and also sells testing and monitoring tools for voice agents, so it has a commercial interest in this market. The page gives no cohort size or measurement date; retrieval date is not the measurement date.
MAI-Voice-2.1 p90 speech-start latency (Show measurement, test conditions, and source)
- Source value
- 1.32
- Score
- 1.32
- Metric
- Time to start speaking for the slowest one in ten sentences, p90
- Unit
- seconds
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Cekura Bench
- Status
- active
- Retrieved at
- 2026-10-23
- Methodology
- Independent evaluation. Cekura reports MAI-Voice-2.1 speech-start latency for a typical sentence and the slowest one in ten. The page does not state a measurement date, utterance count, or immutable serving revision.
- Context
- Cekura says it runs these benchmarks itself and also sells testing and monitoring tools for voice agents, so it has a commercial interest in this market. The page gives no cohort size or measurement date; retrieval date is not the measurement date.
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
20 of 83 models
pyannote Community-1, pyannote: 73
Nemotron-3-Diarization, NVIDIA: 57
Gemini 3.5 Transcribe, Google: 52
Confucius4-R2T2, NetEase Youdao: 52
Granite Speech 5.0 470M TurboCTC, IBM: 52
Kokoro-82M, hexgrad: 51
Nova-3, Deepgram: 51
Fish Audio S2 Pro, Fish Audio: 51
NVIDIA Nemotron 3.5 Transcribe, NVIDIA: 50
Granite Speech 5.0 470M TurboCTC NC, IBM: 50
Orukeet, Oruk: 49
Lyria 3.5, Google: 47
Gemini 3.5 Live Translate, Google: 46
Phonon-2, Fermion Research: 44
Aura-2, Deepgram: 44
Gemini 3.1 Flash TTS, Google: 44
MiniMax-Music3, MiniMax: 41
Audio8-ASR-Infinite, Edge0 AI: 41
GPT-Realtime 2.1, OpenAI: 41
MAI-Voice-2.1, Microsoft: 1
MAI-Voice-2.1
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 5, 2026
MAI-Voice-2.1Qwen3.8-LiveTranslateEleven v4MAI-Transcribe-2-StreamingCartesia Ink-2
All specifications
Architecture
Input and output
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
API
Further details
Prices
- Price
- $22 per 1M charactersSource
Sources and data date
Show all 11 sources
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.