Is this page helpful?

Voices and Emotional Styles#

Each TTS model ships with a catalog of built-in voices. This page explains how voice names are structured, which speakers support emotional style variants, and how to discover and select voices at runtime.

Naming Convention#

Voice names follow a hierarchical pattern that encodes the model, locale, speaker, and optional emotional style.

Magpie TTS Multilingual#

Magpie-Multilingual.{LOCALE}.{Speaker}[.{Emotion}]

Segment

Description

Examples

Magpie-Multilingual

Model prefix

Always Magpie-Multilingual

{LOCALE}

Language and region in uppercase

EN-US, ES-US, FR-FR, DE-DE, ZH-CN, VI-VN, IT-IT, HI-IN, JA-JP, KO-KR, AR-AR, PT-BR

{Speaker}

Speaker identity

Aria, Diego, HouZhen, Isabela, Jason, Leo, Long, Louise, Mia, Pascal, Ray, Siwei, Sofia

{Emotion}

Optional emotional style

Angry, Calm, Fearful, Happy, Neutral, Sad, PleasantSurprised

When you omit the emotion segment (for example, Magpie-Multilingual.EN-US.Aria), the model uses a default neutral style for that speaker.

Examples:

  • Magpie-Multilingual.EN-US.Aria – Aria with default style

  • Magpie-Multilingual.EN-US.Aria.Angry – Aria with angry emotional style

  • Magpie-Multilingual.FR-FR.Pascal.Calm – Pascal (French) with calm style

  • Magpie-Multilingual.VI-VN.Long.Sad – Long (Vietnamese) with sad style

Magpie TTS Zeroshot#

Magpie-ZeroShot-Multilingual[.{Gender}]

Built-in voices use Magpie-ZeroShot-Multilingual.Female and Magpie-ZeroShot-Multilingual.Male. For zero-shot voice cloning from an audio prompt, use Magpie-ZeroShot-Multilingual without a gender suffix. The same voices are available across all supported languages.

Examples: Magpie-ZeroShot-Multilingual.Female, Magpie-ZeroShot-Multilingual.Male, Magpie-ZeroShot-Multilingual (with an audio prompt)

Magpie TTS Flow#

English-US-Magpie-Flow.{Gender}[.{Emotion}]

Similar to Zeroshot, but uses dots to separate the emotion. Default voices use {Gender}-1.

Examples: English-US-Magpie-Flow.Female-1, English-US-Magpie-Flow.Male.Happy

Emotional Style Availability#

Not all speakers have emotional variants, and the available emotions differ by speaker and locale.

Emotions by Model#

Emotion

Multilingual

Zeroshot

Flow

Angry

✅

❌

✅

Calm

✅

❌

✅

Fearful

✅

❌

✅

Happy

✅

❌

✅

Neutral

✅

❌

✅

Sad

✅

❌

✅

Disgusted

❌

❌

✅

PleasantSurprised

✅

❌

❌

Multilingual Speakers with Emotional Variants#

In the Magpie TTS Multilingual model, emotional styles are available only for specific speaker-locale combinations.

Speaker

EN-US

ES-US

FR-FR

DE-DE

ZH-CN

VI-VN

IT-IT

HI-IN

JA-JP

KO-KR

AR-AR

PT-BR

Aria

4 emotions

–

–

–

–

–

–

–

–

default only

–

–

Jason

4 emotions

–

–

default only

–

default only

–

–

–

–

default only

–

Leo

4 emotions

–

–

default only

–

–

–

default only

–

–

–

–

Mia

4 emotions

–

–

default only

–

–

–

–

–

–

–

–

Ray

4 emotions

–

–

default only

–

–

–

–

default only

default only

default only

–

Sofia

4 emotions

–

–

–

–

–

–

default only

–

–

default only

–

Diego

–

5 emotions

–

default only

–

–

–

–

–

default only

–

default only

Isabela

–

5 emotions

–

–

–

default only

default only

–

default only

–

–

default only

Pascal

–

–

4 emotions

default only

–

default only

default only

default only

–

default only

–

–

Long

–

–

–

–

–

3 emotions

–

–

–

–

–

–

Speakers without emotional variants (for example, Pascal in DE-DE and Louise in FR-FR) produce speech in a default neutral style.

Discover Voices at Runtime#

List the voices a deployed TTS NIM is currently serving.

python3 python-clients/scripts/tts/talk.py \
    --server 0.0.0.0:50051 \
    --list-voices

curl -sS http://localhost:9000/v1/audio/list_voices | jq

python3 python-clients/scripts/tts/realtime_tts_client.py \
    --server localhost:9000 \
    --list-voices

The output lists voice names grouped by language code. Use these exact names in the --voice parameter when synthesizing speech.

Select a Voice#

Pass the full voice name to the --voice flag. If omitted, the server selects the first available voice for the specified --language-code.

python3 python-clients/scripts/tts/talk.py --server 0.0.0.0:50051 \
    --language-code en-US \
    --text "Hello from Aria with a calm tone." \
    --voice Magpie-Multilingual.EN-US.Aria.Calm \
    --output output.wav

Cross-Language Accents#

The Magpie TTS Multilingual model allows mixing a voice from one locale with text in another language to produce accented speech. For example, synthesize English text with a French-accented voice:

python3 python-clients/scripts/tts/talk.py --server 0.0.0.0:50051 \
    --language-code en-US \
    --text "This English text is spoken with a French accent." \
    --voice Magpie-Multilingual.FR-FR.Pascal \
    --output output.wav

Complete Voice Reference#

For the full list of all available voices per model and locale, refer to the TTS support matrix.