Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Microsoft AI / MAI-Voice-2.1

MAI-Voice-2.1

MAI-Voice-2.1 converts text to expressive speech with curated voices, SSML style controls and gated reference-voice cloning. The standard variant supports consistent voices for long-form narration.

Speech & voice
previewClosed weightsText-to-speech
Official documentation (opens in a new tab)Reviewed 14 Oct 2026
10 of 23 specification fields collected
Output formatsMP3See specification notes below
Languages28 listedFull language list in specifications
Input billingSee ratesAzure Speech
Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected
Maximum input tokens
Not collected
Maximum output
Not collected
Reasoning
Not collected
Tool calling
Not collected
Structured output
Not collected
Streaming
Not collected
Batch processing
Not collected
Fine-tuning
Not collected
Knowledge cutoff
Not collected
Release date
Not collected
Parameters
Not collected
Languages
View all 28 entries
  • cs-CZ
  • da-DK
  • de-DE
  • en-AU
  • en-GB
  • en-IN
  • en-US
  • es-ES
  • es-MX
  • fi-FI
  • fr-FR
  • hi-IN
  • hu-HU
  • id-ID
  • it-IT
  • ko-KR
  • nb-NO
  • nl-NL
  • pl-PL
  • pt-BR
  • pt-PT
  • ro-RO
  • ru-RU
  • sv-SE
  • th-TH
  • tr-TR
  • vi-VN
  • zh-CN
Source for Languages (opens in a new tab)

28 locale variants in the reviewed voice catalog. Microsoft describes coverage as 23 languages.

Context caching
Not collected
Input modalities
TextSource for Input modalities (opens in a new tab)

Reference audio is additionally used for gated voice cloning.

Output modalities
AudioSource for Output modalities (opens in a new tab)
Input file formats
SSMLSource for Input file formats (opens in a new tab)
Output formats
MP3Source for Output formats (opens in a new tab)

Verified REST example uses audio-24khz-160kbitrate-mono-mp3; this is not asserted to be an exhaustive format list.

Stock voice catalog
97 named voice/locale entries; individual voices have separate supported style lists.Source for Stock voice catalog (opens in a new tab)
Voice controls
SSML voice selection and mstts:express-as style. Available styles depend on the voice.Source for Voice controls (opens in a new tab)
Voice cloning
SupportedSource for Voice cloning (opens in a new tab)

Gated access; a consented reference clip is required. This is not evidence of fine-tuning support.

Voice reference requirements
5–60 seconds of consented reference audio, without additional training; gated availability.Source for Voice reference requirements (opens in a new tab)
Provider-reported latency
Approximately 550 ms model inference latency.Source for Provider-reported latency (opens in a new tab)

Provider-reported model inference figure, not an end-to-end latency guarantee or independently reproduced measurement.

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Azure Speech

Provider source (opens in a new tab)
API model ID
MAI-Voice-2.1

Select the model through a voice identifier such as en-US-Harper:MAI-Voice-2.1 in SSML, not a JSON model parameter.

Access
POST https://{region}.tts.speech.microsoft.com/cognitiveservices/v1 using SSML, or Azure Speech SDK. Voice Live integration is also documented; its combined session pricing is not inferred from TTS rates.
Regions
Globally accessible; documented serving regions: francecentral, eastasia, southeastasia, eastus, canadacentral, eastus2, westus, westeurope, northeurope, westus2, westus3, centralindia, swedencentral, japaneast.

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 14 Oct 2026
CategoryUsageRateUnit
Speech synthesisText-to-speech characters1$22.001M characters

Rate conditions

  1. Published USD rate for this exact model version. Do not interpret characters as tokens. Voice Live orchestration, reference-voice enrollment and other service charges are not included in this standalone synthesis rate.
Pricing source (opens in a new tab)

Data & enterprise terms

Model-specific retention and training-use terms not collected.

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

Azure Speech

MAI-Voice-2.1
Not verified

No model-specific cache policy verified for this endpoint.

How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 14 Oct 2026

Voice choices

97 voices

Choose a voice within MAI-Voice-2.1. These are API voice identifiers, not additional model releases. Source links document each voice.

Published voice catalog
Voice / API IDLanguage & accentDelivery
Grantcs-CZ-Grant:MAI-Voice-2.1Czech
cs-CZ		Male. Supported styles: agent, audiobook, customer_call_center, educational, narrator, neutral.Voice source (opens in a new tab)
Harpercs-CZ-Harper:MAI-Voice-2.1Czech
cs-CZ		Female. Supported styles: agent, audiobook, customer_call_center, educational, narrator, neutral.Voice source (opens in a new tab)
Grantda-DK-Grant:MAI-Voice-2.1Danish
da-DK		Male. Supported styles: agent, customer_call_center, educational, narrator, neutral.Voice source (opens in a new tab)
Harperda-DK-Harper:MAI-Voice-2.1Danish
da-DK		Female. Supported styles: agent, audiobook, customer_call_center, educational, narrator, neutral.Voice source (opens in a new tab)
Grantde-DE-Grant:MAI-Voice-2.1German
de-DE		Male. Supported styles: agent, audiobook, customer_call_center, educational, narrator, neutral.Voice source (opens in a new tab)
Harperde-DE-Harper:MAI-Voice-2.1German
de-DE		Female. Supported styles: agent, audiobook, customer_call_center, educational, narrator, neutral.Voice source (opens in a new tab)
Klausde-DE-Klaus:MAI-Voice-2.1German (Germany)
de-DE		Male. Supported styles: angry, confused, determined, disgusted, embarrassed, excited, fearful, happy, hopeful, jealous, joyful, neutral, regretful, relieved, sad, shouting, softvoice, surprised, whispering.Voice source (opens in a new tab)
Miade-DE-Mia:MAI-Voice-2.1German (Germany)
de-DE		Female. Supported styles: angry, confused, determined, disgusted, embarrassed, excited, fearful, happy, hopeful, jealous, joyful, neutral, regretful, relieved, sad, shouting, softvoice, surprised, whispering.Voice source (opens in a new tab)

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Microsoft identifies this as its own MAI model, supplied through managed inference endpoints. The reviewed model documentation provides no downloadable model weights. This record does not classify public Phi or MAI-DS-R1 weights as closed.

Classification evidence (opens in a new tab)

Sources reviewed