{"id":2887,"date":"2026-09-10T11:08:42","date_gmt":"2026-09-10T18:08:42","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/foundry\/?p=2887"},"modified":"2026-09-10T11:22:32","modified_gmt":"2026-09-10T18:22:32","slug":"announcing-azure-ai-speech-llm-2607","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/foundry\/announcing-azure-ai-speech-llm-2607\/","title":{"rendered":"Announcing Azure AI Speech LLM 2607: Better Multilingual Accuracy, Easier Customization"},"content":{"rendered":"<p>Azure AI Speech continues to deliver regular improvements to LLM Speech throughout the year, helping developers build more accurate and capable voice experiences. We\u2019re announcing <strong>LLM Speech 2607<\/strong>, our latest model update, which brings improvements in multilingual recognition and mixed-language audio inputs, along with an enhanced experience for the phrase list feature.<\/p>\n<p>Whether you\u2019re building voice agents, meeting transcription solutions, contact center experiences, or online meeting applications, LLM Speech 2607 delivers transcripts you can trust for high-stakes, high-volume communication with no changes to your application architecture.<\/p>\n<p>Try it out now in the <a href=\"https:\/\/ai.azure.com\/?cid=learnDocs\">Foundry Playground<\/a> (Azure Speech \u2013 Speech to Text), or see REST API or SDK modalities in our <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-services\/speech-service\/fast-transcription-create?tabs=new-foundry%2Clocale-specified%2Cwindows&amp;pivots=programming-language-rest\">public docs<\/a>.<\/p>\n<h2>Key Advancements in Model Quality and Accuracy<\/h2>\n<p>This release delivers a decisive quality leap across Tier 1 and Tier 2 languages.<\/p>\n<p>Building on the previous 2605 release, our internal evaluations also show that LLM Speech 2607 delivers measurable quality improvements across key scenarios:<\/p>\n<ul>\n<li><strong>Mixed-language audio<\/strong>, where the language spoken changes mid-conversation, improving recognition when speakers switch languages during a conversation.<\/li>\n<li><strong>Improved accuracy on Tier 2 and Tier 3 language locales<\/strong>, helping close the quality gap across a broader set of languages. In particular, capitalization, punctuation, and numeric and digit transcription are more robust.<\/li>\n<li><strong>Domain-aware entity and person-name recognition<\/strong>, precisely capturing names, brands, and domain technology.<\/li>\n<li><strong>Lower latency<\/strong>, improved up to 3x faster than the previous 2605 release. This is especially relevant for crisp, real-time responsiveness for interactive use.<\/li>\n<\/ul>\n<p>Screenshot below shows a comparison of the previous 2605 release and the new 2607 release on two sample audio inputs. Punctuation and numeric transcription improvements are highlighted.<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/devblogs.microsoft.com\/foundry\/wp-content\/uploads\/sites\/89\/2026\/09\/2607-llm-sppech-transcription-quality-comparision.webp\" alt=\"Side-by-side comparison of 2605 and 2607 model quality\" \/><\/p>\n<h2>New Enhancement: Phrase Lists<\/h2>\n<p>A common challenge in speech applications is accurately recognizing domain-specific vocabulary such as product names, technical terminology, acronyms, company names, or industry jargon.<\/p>\n<p>With LLM Speech 2607, we\u2019re simplifying how a customer-provided phrase list can be enabled: use it as a dedicated parameter for providing important words and phrases instead of embedding custom vocabulary into a catch-all prompt. You can now specify recognition hints directly through a purpose-built field.<\/p>\n<p>This makes customization simpler, cleaner, and easier to maintain in production applications. Phrase lists can include 2,000+ entities, providing a significant quality boost and allowing the LLM to be more context-aware. Further documentation can be found <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-services\/speech-service\/improve-accuracy-phrase-list?tabs=terminal&amp;pivots=programming-language-csharp\">here<\/a>.<\/p>\n<h3>Example request including phrase list<\/h3>\n<pre><code class=\"language-bash\">curl --location 'https:\/\/YourResourceName.cognitiveservices.azure.com\/speechtotext\/transcriptions:transcribe?api-version=2025-10-15' \\\r\n  --header 'Ocp-Apim-Subscription-Key: YourSpeechResourceKey' \\\r\n  --form 'audio=@\"YourAudioFile\"' \\\r\n  --form 'definition={\r\n    \"locales\": [\"en-US\"],\r\n    \"phraseList\": {\r\n      \"phrases\": [\"Contoso\", \"Jessie\", \"Rehaan\"]\r\n    }\r\n  }'<\/code><\/pre>\n<p>Phrase Lists are especially useful for scenarios involving:<\/p>\n<ul>\n<li>Enterprise terminology<\/li>\n<li>Technical product names<\/li>\n<li>Industry-specific vocabulary<\/li>\n<li>Acronyms and abbreviations<\/li>\n<li>Customer-defined terms<\/li>\n<\/ul>\n<h2>Available through Fast API and Real-Time API<\/h2>\n<p>LLM Speech 2607 is available through the <strong>Fast API<\/strong>, enabling developers to bring the latest model improvements to both transcription and interactive voice experiences.<\/p>\n<h2>Get started<\/h2>\n<p>LLM Speech 2607 is the latest step in our ongoing investment to make speech AI more accurate, multilingual, and easier to customize. No customer action is required; this update will be deployed automatically on the service side. Try it out now in the <a href=\"https:\/\/ai.azure.com\/?cid=learnDocs\">Foundry Playground<\/a> (Azure Speech \u2013 Speech to Text), or see REST API or SDK modalities in our <a href=\"https:\/\/learn.microsoft.com\/en-us\/azure\/ai-services\/speech-service\/fast-transcription-create?tabs=new-foundry%2Clocale-specified%2Cwindows&amp;pivots=programming-language-rest\">public docs<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Learn how Azure AI Speech LLM 2607 improves multilingual and mixed-language recognition, lowers latency, and simplifies domain-specific customization with enhanced phrase lists.<\/p>\n","protected":false},"author":202606,"featured_media":2888,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1],"tags":[177,178,2,180,181,179],"class_list":["post-2887","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-microsoft-foundry","tag-azure-ai-speech","tag-llm-speech","tag-microsoft-foundry","tag-multilingual-speech","tag-phrase-lists","tag-speech-to-text"],"acf":[],"blog_post_summary":"<p>Learn how Azure AI Speech LLM 2607 improves multilingual and mixed-language recognition, lowers latency, and simplifies domain-specific customization with enhanced phrase lists.<\/p>\n","_links":{"self":[{"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/posts\/2887","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/users\/202606"}],"replies":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/comments?post=2887"}],"version-history":[{"count":1,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/posts\/2887\/revisions"}],"predecessor-version":[{"id":2891,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/posts\/2887\/revisions\/2891"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/media\/2888"}],"wp:attachment":[{"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/media?parent=2887"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/categories?post=2887"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/foundry\/wp-json\/wp\/v2\/tags?post=2887"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}