Deepgram's Flux STT cuts unwanted interruptions in Anam's avatar conversations by ~50%
In short: Anam, which builds photorealistic interactive avatars, switched to Deepgram's Flux speech-to-text model to solve turn-taking problems in unscripted, real-time conversation. In A/B testing against a self-hosted open-source STT option, Flux cut unwanted avatar barge-ins by roughly 50% with no added latency penalty, and complaints about turn-taking on English sessions dropped to near zero. Anam deployed Flux self-hosted on its own Kubernetes/H100 infrastructure, rolling it out to 100% of turnkey users after a two-week self-hosted deployment (versus days for Deepgram's hosted option).
This summary was generated automatically by AI from Deepgram's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1~50% fewer unwanted avatar barge-ins in A/B testing versus a self-hosted open-source STT baseline
- 2p95 latency stayed flat ('0 latency penalty') compared to tested alternatives
- 3Turn-taking complaints on English sessions dropped to near zero after rollout
- 4Self-hosted production deployment took about two weeks on Anam's own Kubernetes cluster with H100 GPUs; Deepgram-hosted deployment took only days
- 5Rolled out to 100% of Anam's turnkey users after validation
- 6Anam is also evaluating Deepgram's Flux TTS offering for future voice quality improvements
Why it matters
Turn-taking — knowing when a speaker has actually finished versus just pausing — is one of the hardest unsolved problems in real-time conversational AI, and getting it wrong breaks the illusion of a natural conversation immediately. This case study shows a purpose-built STT model can meaningfully reduce interruptions without adding latency, and that it can be self-hosted for latency-sensitive, high-concurrency deployments rather than relying only on a cloud API.
What it means for AI agents and contact centers
For anyone building phone-based voice agents, this is a direct data point on how a turn-taking-aware STT model performs against a self-hosted open-source baseline in production, including deployment timelines (days hosted vs. two weeks self-hosted) and concrete metrics like barge-in reduction and flat p95 latency. Worth comparing against current STT providers on turn-taking accuracy, resilience to background noise, and concurrency scaling, especially if self-hosting is a requirement for latency or data control.
🧪 Worth Testing
Flux's documented reduction in unwanted interruptions and stable latency under real-world A/B testing makes it a strong candidate to benchmark against current STT pipelines for call-based voice agents, particularly where natural turn-taking and self-hosting are priorities.
Sources
- DeepgramOfficialPrimary sourceOriginal article →„Article · · AI Agents Anam Solves Turn-Taking for Unscripted Avatar Conversation with Flux STT“14 Oct 2026, 03:00
- Published by source
- 14 Oct 2026, 03:00
- Found by our system
- 15 Oct 2026, 16:11
- Summary generated
- 15 Oct 2026, 16:13
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy