--- title: How to Evaluate Live & Voice Agents in ADK | daily.dev description: Google's Agent Development Kit (ADK) now supports native live evaluation for voice-based agents, letting developers drive a live agent with a simulated audio... canonical: https://daily.dev/posts/how-to-evaluate-live-voice-agents-in-adk-mx6s58g3a twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: How to Evaluate Live & Voice Agents in ADK | daily.dev og:description: Google's Agent Development Kit (ADK) now supports native live evaluation for voice-based agents, letting developers drive a live agent with a simulated audio... og:url: https://daily.dev/posts/how-to-evaluate-live-voice-agents-in-adk-mx6s58g3a og:image: https://api.daily.dev/og/posts/Mx6s58g3A.png og:image:alt: How to Evaluate Live & Voice Agents in ADK og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # How to Evaluate Live & Voice Agents in ADK **[Google Developers](https://daily.dev/sources/googledevs)** · 6 min read · 3 upvotes · 1 comments ## Summary Google's Agent Development Kit (ADK) now supports native live evaluation for voice-based agents, letting developers drive a live agent with a simulated audio user, score multi-turn spoken conversations with rubric-based judging, and run the same eval loop used for text agents. The walkthrough builds a three-stage graph-based live workflow (greeter, DOB verifier, goals agent) running on gemini-live-2.5-flash-native-audio, shows how to author conversation-scenario and fixed-conversation eval cases, configure a user simulator with Gemini TTS voices, run evals via CLI or programmatically through AgentEvaluator for CI/CD, and inspect results with playable audio transcripts in ADK Web. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How do I evaluate a live voice agent built with Google ADK? ADK now includes native live evaluation: you drive a live agent with a simulated audio user, score its spoken replies with rubric-based multi-turn trajectory criteria, and run it via the same eval loop used for text agents. Configure a test_config.json with an llm_audio user_simulator_config pointing to a model and audio_model, then run 'adk eval' from the CLI or call AgentEvaluator programmatically for CI/CD. _daily.dev surfaces workflow guides like this for teams automating voice agent regression testing._ ### What is the difference between a conversation scenario and a fixed conversation in an ADK eval set? A conversation scenario gives the user simulator a starting prompt, a conversation_plan, and a user_persona (such as NOVICE), letting it improvise turns and end the scenario once the plan is satisfied, capped by max_allowed_invocations. A fixed conversation instead scripts the user's turns verbatim as text, giving a static, reproducible input to the same live evaluation run. _developers designing agent test suites can track ADK eval patterns like these on daily.dev._ ### What models power ADK's live voice agent evaluation user simulator? The example configuration uses gemini-3.7-flash as the judge_model and simulator model for turn-taking logic, and gemini-3.1-flash-tts-preview as the audio_model to synthesize the simulated user's turns into speech, with Gemini TTS voices like 'Kore' configurable via voice_config and language_code. _anyone comparing Gemini model options for voice agent pipelines can follow updates like these on daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@kartiknvj** · 0 upvotes > The inline audio playback next to the transcript is the piece I wish every voice eval tool had, because reading a transcript will not tell you the agent talked over the user or dragged the pause. I have found improvised scenarios catch far more real bugs than fixed conversations, but only if the persona actually pushes back instead of politely accepting whatever the agent says. Are the natural language rubrics scored by a judge model, and if so how are you calibrating drift on it over time? ## Similar posts on daily.dev - [Announcing User Simulation in ADK Evaluation](https://daily.dev/posts/announcing-user-simulation-in-adk-evaluation-hgcgnjkrw) · Google Developers · 0 upvotes · 0 comments - [13 demos on Gemini Enterprise Agent Platform](https://daily.dev/posts/13-demos-on-gemini-enterprise-agent-platform-sxyv8lnfw) · Google Cloud · 1 upvotes · 0 comments - [Hands-on with the Google Agent Development Kit](https://daily.dev/posts/hands-on-with-the-google-agent-development-kit-zcvlgqadp) · InfoWorld · 1 upvotes · 0 comments - [Karpathy's Agentic Engineering Finally Has Proper Tooling](https://daily.dev/posts/karpathy-s-agentic-engineering-finally-has-proper-tooling-wm38wjfvk) · Daily Dose of Data Science \| Avi Chawla \| Substack · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#cicd](https://daily.dev/tags/cicd), [#google-gemini](https://daily.dev/tags/google-gemini), [#voice-ai](https://daily.dev/tags/voice-ai) [View this post on daily.dev](https://daily.dev/posts/how-to-evaluate-live-voice-agents-in-adk-mx6s58g3a) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"How to Evaluate Live & Voice Agents in ADK","url":"https://daily.dev/posts/how-to-evaluate-live-voice-agents-in-adk-mx6s58g3a","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/how-to-evaluate-live-voice-agents-in-adk-mx6s58g3a"},"datePublished":"2026-08-24T18:13:37.995Z","dateModified":"2026-09-14T08:46:28.562Z","description":"Google's Agent Development Kit (ADK) now supports native live evaluation for voice-based agents, letting developers drive a live agent with a simulated audio...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c53ca65c20fe259770709bab56a15f33?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c53ca65c20fe259770709bab56a15f33?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Google Developers","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Google Developers","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/ea4bbded4e7f45ccb82e121a3156535f","url":"https://daily.dev/sources/googledevs"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/how-to-evaluate-live-voice-agents-in-adk-mx6s58g3a","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"llm,cicd,google-gemini,voice-ai","timeRequired":"PT6M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Google Developers","item":"https://daily.dev/sources/googledevs"},{"@type":"ListItem","position":3,"name":"How to Evaluate Live & Voice Agents in ADK"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/how-to-evaluate-live-voice-agents-in-adk-mx6s58g3a","comment":[{"@type":"Comment","text":"The inline audio playback next to the transcript is the piece I wish every voice eval tool had, because reading a transcript will not tell you the agent talked over the user or dragged the pause. I have found improvised scenarios catch far more real bugs than fixed conversations, but only if the persona actually pushes back instead of politely accepting whatever the agent says. Are the natural language rubrics scored by a judge model, and if so how are you calibrating drift on it over time?","datePublished":"2026-08-25T20:30:08.432Z","url":"https://daily.dev/posts/Mx6s58g3A#c-zJwmgOlIE","author":{"@type":"Person","name":"kartik-nvjk","url":"https://daily.dev/kartiknvj","image":"https://media.daily.dev/image/upload/s--3gGgsVCw--/f_auto/v1781456774/avatars/avatar_TvTVeiMdkRCqWUDullFmy?_a=BAMAMiWQ0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/how-to-evaluate-live-voice-agents-in-adk-mx6s58g3a#faq","mainEntity":[{"@type":"Question","name":"How do I evaluate a live voice agent built with Google ADK?","acceptedAnswer":{"@type":"Answer","text":"ADK now includes native live evaluation: you drive a live agent with a simulated audio user, score its spoken replies with rubric-based multi-turn trajectory criteria, and run it via the same eval loop used for text agents. Configure a test_config.json with an llm_audio user_simulator_config pointing to a model and audio_model, then run 'adk eval' from the CLI or call AgentEvaluator programmatically for CI/CD. daily.dev surfaces workflow guides like this for teams automating voice agent regression testing."}},{"@type":"Question","name":"What is the difference between a conversation scenario and a fixed conversation in an ADK eval set?","acceptedAnswer":{"@type":"Answer","text":"A conversation scenario gives the user simulator a starting prompt, a conversation_plan, and a user_persona (such as NOVICE), letting it improvise turns and end the scenario once the plan is satisfied, capped by max_allowed_invocations. A fixed conversation instead scripts the user's turns verbatim as text, giving a static, reproducible input to the same live evaluation run. developers designing agent test suites can track ADK eval patterns like these on daily.dev."}},{"@type":"Question","name":"What models power ADK's live voice agent evaluation user simulator?","acceptedAnswer":{"@type":"Answer","text":"The example configuration uses gemini-3.7-flash as the judge_model and simulator model for turn-taking logic, and gemini-3.1-flash-tts-preview as the audio_model to synthesize the simulated user's turns into speech, with Gemini TTS voices like 'Kore' configurable via voice_config and language_code. anyone comparing Gemini model options for voice agent pipelines can follow updates like these on daily.dev."}}]} ```