MAI-Voice-1#

Synthesize speech with Azure AI Speech using Microsoft's MAI-Voice-1 voices.

Quick start#

bash

{baseDir}/scripts/speak.sh --text "Hello Steve"

Defaults:

• Voice: en-us-Jasper:MAI-Voice-1
• Output: ./mai-voice.mp3
• Format: audio-24khz-160kbitrate-mono-mp3

Useful flags#

bash

{baseDir}/scripts/speak.sh --text "Hello Steve" --voice en-us-Iris:MAI-Voice-1 --out /tmp/iris.mp3
{baseDir}/scripts/speak.sh --text-file /tmp/input.txt --voice en-us-June:MAI-Voice-1 --style empathy --out /tmp/june.mp3
{baseDir}/scripts/speak.sh --text "Let's go" --voice en-us-Jasper:MAI-Voice-1 --style excitement
{baseDir}/scripts/speak.sh --list-voices

Required env vars#

bash

export AZURE_SPEECH_KEY="YOUR_SPEECH_RESOURCE_KEY"
export AZURE_SPEECH_REGION="eastus"

How to get the API key and region#

1. Go to the Azure portal and open your Speech or Foundry Speech resource.
2. Open Keys and Endpoint.
3. Copy one of the resource keys.
4. Copy the resource region, for example eastus.
5. Export them:

bash

export AZURE_SPEECH_KEY="YOUR_SPEECH_RESOURCE_KEY"
export AZURE_SPEECH_REGION="eastus"

The MAI-Voice docs currently point at East US for preview access, so if you are not using eastus, double-check that your resource actually supports the model before blaming the script.

Optional:

bash

export AZURE_SPEECH_OUTPUT_FORMAT="audio-24khz-160kbitrate-mono-mp3"

Supported voices#

• en-us-Jasper:MAI-Voice-1
• en-us-June:MAI-Voice-1
• en-us-Grant:MAI-Voice-1
• en-us-Iris:MAI-Voice-1
• en-us-Reed:MAI-Voice-1
• en-us-Joy:MAI-Voice-1

API shape#

The script calls:

text

POST https://{AZURE_SPEECH_REGION}.tts.speech.microsoft.com/cognitiveservices/v1

Headers:

• Ocp-Apim-Subscription-Key: {AZURE_SPEECH_KEY}
• Content-Type: application/ssml+xml
• X-Microsoft-OutputFormat: {format}
• User-Agent: curl

Body:

• SSML with a MAI voice name
• optional mstts:express-as style wrapper

Notes#

• This uses the Azure Speech REST API, not the Python SDK.
• Voice selection is user-configurable.
• Style is optional and applied via SSML.
• MAI-Voice-1 is currently public preview.