Microsoft Launches Dragon HD Omni: 700+ Multilingual Azure Voices With New Styles in Preview Right Now

Microsoft is rolling out a major upgrade to Azure Speech with a new text-to-speech voice type called Dragon HD Omni, now in preview via Microsoft Foundry and initially available in East US, West Europe, Sweden Central, and Southeast Asia regions. Built as a unified generation model, Dragon HD Omni brings over 700 upgraded and new voices, richer expressiveness, and deep multilingual support, all designed to dramatically reduce SSML tuning and help developers ship more natural voice experiences faster.​

Dragon HD Omni: A New Unified Voice Generation Model

Dragon HD Omni is the latest generation of Azure Speech text-to-speech, built as a unified model that powers more than 700 high-quality voices across a wide range of languages and personas. Instead of juggling separate models or voice families, developers can now tap into a single model designed to deliver better prosody, context awareness, and expressiveness out of the box.​

Microsoft positions Dragon HD Omni as a direct answer to common developer pain points in production TTS: robotic prosody, narrow language coverage, and the need for heavy SSML micromanagement just to get speech to sound natural. By consolidating generations and upgrading existing voices into the Omni model, Azure Speech aims to make natural-sounding, production-ready speech the default rather than something you have to fight to achieve.​

700+ Voices, Including 300 New AI‑Generated Personas

With Dragon HD Omni, almost all of Azure’s familiar neural voices are upgraded and merged into a single, richer model while preserving each voice’s unique character. Voice names now follow a consistent pattern, where you can access the new generation simply by appending the :DragonHDOmniLatestNeural suffix to an existing voice, such as en-US-Ava:DragonHDOmniLatestNeural or en-US-Andrew:DragonHDOmniLatestNeural.​

On top of the upgraded catalog, Dragon HD Omni introduces nearly 300 brand-new AI‑generated voices designed to broaden vocal diversity for modern applications. These new voices span:​

• Gender: male, female, and non-binary options for more inclusive design.​

• Age: youthful, mature, and senior tonalities to better fit scenarios like children’s content vs. corporate narration.​

• Pitch and tone: from warm and conversational to formal, authoritative, or highly dramatic delivery.​

The idea is to give teams enough variety to:

• Personalize experiences per audience segment, whether that’s education, support, productivity, or entertainment.​

• Build stronger brand identity by consistently using voices that align with brand values and personality.​

• Increase inclusivity by representing more vocal styles and identities across cultures and regions.​

Microsoft is already highlighting new persona-style voices with descriptive labels such as a bold, dramatic male voice (“en-us-graphiterhodium”) or a calm, soothing adult female (“en-us-olivepoivre”), giving creators a clearer mental model of how each voice will sound before they integrate it.​

Style Control: Natural Language Styles Instead of Heavy SSML

Traditional Azure neural voices supported only a limited set of styles, and getting them to sound “just right” often required painstaking SSML tuning. Dragon HD Omni shifts that model by adding automatic style prediction powered by natural language descriptions, significantly lowering the friction for experimenting with expressive voices.​

In the initial preview, style control is rolling out for the popular en-US-Ava and en-US-Andrew voices, but the roadmap clearly points toward broader coverage. Developers can specify styles such as “angry”, “friendly”, “news”, “narration”, “emo teenager”, “mad scientist”, “meditative”, “santa”, “soft voice”, and many more, with the model interpreting and applying these cues directly. Because style is strongly influenced by the input text, the same style label can produce different nuances depending on the script, making it easier to blend emotion and context naturally.​

From a practical standpoint, this style system lets teams:

• Build more engaging virtual agents and IVRs that move beyond a single neutral tone.​

• Produce podcasts, audiobooks, or marketing content with distinct voices per segment or character.​

• Rapidly prototype and A/B test multiple emotional deliveries without rewriting SSML trees.​

Multilingual Voices and Accent Flexibility

Another major pillar of Dragon HD Omni is deep multilingual support. All Omni voices can handle multiple languages, automatically predicting and generating speech in the language that matches the input text. That means a single voice can fluidly switch from, say, English to French to German in one script, a scenario that previously required more manual voice juggling.​

For precise control, developers can still use SSML’s <lang> element to adjust speaking languages and accents, specifying tags such as fr-FR for French or de-DE for German. This pattern is especially valuable for:​

• Language-learning apps that mix explanations in one language with examples in another.

• Travel and navigation scenarios where localized phrases matter for user trust.

• International business content that needs native-sounding pronunciation across markets.

Under the hood, these multilingual capabilities build on Azure Speech’s existing support for more than 150 languages and locales and hundreds of neural voices, but Dragon HD Omni adds a new level of contextual switching and prosody control for cross-lingual content.​

Word Boundary Events for Karaoke, Captions, and Sync

Beyond sound quality and voice variety, Dragon HD Omni introduces precise word boundary event support, enabling developers to track the exact timing of each word in the synthesized audio. As the engine speaks, it emits events that include the text of the word, an audio offset in milliseconds, and the position of that word in the original input text.​

This capability unlocks several high-value scenarios:

• Karaoke-style experiences where highlighted words stay perfectly in sync with the audio.​

• Real-time captioning and accessibility overlays that follow spoken content with word-level precision.​

• Interactive storytelling or learning apps where clicking or hovering on text jumps to the corresponding audio moment.

Microsoft’s sample code shows how to wire this up in Python using the Azure Speech SDK: you register a callback on synthesis_word_boundary, then inspect evt.text, evt.audio_offset, and evt.text_offset for each word as it’s spoken. This gives developers low-level control while still relying on the high-level TTS service for quality and language handling.​

Fine-Tuning Output With New Parameters

Dragon HD Omni exposes several tuning parameters that let teams trade off between stability, creativity, and speed depending on their scenario. The key parameters are:​

• temperature (default 0.7, range 0.3–1.0): controls how varied or creative the speech delivery is. Lower values yield predictable, consistent delivery, while higher values inject more expressive variation.​

• top_p (default 0.7, range 0.3–1.0): filters the set of candidate outputs to encourage diversity; Microsoft recommends keeping this close to temperature for best results.​

• top_k (default 22, range 1–50): limits the number of options considered when generating speech, with higher values typically supporting more varied expression.​

• cfg_scale (default around 1.4, range 1.0–2.0): controls how strongly the model adheres to the input prompt, with higher values generally leading to faster, more context-aligned speech and lower values delivering a slower, more neutral output.​

Microsoft’s guidance is straightforward:

• For more expressive reads, increase temperature, top_p, and top_k together.​

• For maximum stability (e.g., compliance or instruction content), first lower temperature, then adjust top_p if needed.​

• For faster, strongly contextual speech, increase cfg_scale; for slower, more neutral tone, decrease it slightly.​

Developers can pass these as a parameters attribute in SSML on the <voice> tag, either individually or combined, which makes experimenting as simple as tweaking one string and replaying the output.​

How Developers Can Try Dragon HD Omni Today

Dragon HD Omni is available in preview through Microsoft Foundry’s Azure Speech experience, with initial region availability in East US, West Europe, Sweden Central, and Southeast Asia. Developers can get hands-on in a few different ways:​

• Speech Playground in Foundry: The web-based playground lets you pick Omni-powered voices, paste sample scripts, choose styles, and instantly listen to the output without writing code.​

• Voice list and SSML upgrade path: Existing voice users can query the voice list API to find supported Omni variants and then append :DragonHDOmniLatestNeural to their current voice names to upgrade.​

• Direct SDK and REST usage: For production scenarios, Azure Speech’s SDKs (including Python, C#, and JavaScript) as well as REST APIs already support the new voice naming scheme and parameters, making it straightforward to swap in Dragon HD Omni for existing TTS integrations.​

As with other Azure Speech offerings, Dragon HD Omni integrates into the broader portfolio of more than 600 neural voices and over 150 languages and locales, plus Custom Neural Voice for organizations that want to build their own proprietary brand voice alongside the Omni catalog.

That combination of a massive prebuilt library, style control, multilingual fluency, and customizable tuning parameters is aimed squarely at developers and content teams who want to push beyond generic voice overs toward highly tailored, humanlike speech experiences at scale.​

Recent Posts

• Microsoft Reveals Brand-New Copilot Checkout and Special Brand Agents to Turn AI Conversations into Retail Conversions for 2026
• Xbox’s Developer_Direct Returns January 22 with Big Gameplay Reveals for Fable, Forza Horizon 6, and Beast of Reincarnation
• Microsoft Reveals Brand-New Copilot Checkout and Special Brand Agents to Turn AI Conversations into Retail Conversions for 2026
• Microsoft 365 in January 2026: Exciting New AI Features, Copilot Upgrades, and What’s Changing for Your Organization
• Xbox Game Pass January 2026: Star Wars Outlaws, Resident Evil Village, and a Galactic Start to the New Year