The Announcement
Modulate has opened its Velma voice intelligence platform to developers via a public API, extending capabilities previously available only to enterprise customers. The Velma Enterprise API, built on Modulate’s proprietary Ensemble Listening Model (ELM) architecture, delivers real-time analysis of conversational audio across dimensions including emotion, intent, behavioral risk, and compliance signals. The company is positioning Velma not as a transcription tool, but as a continuous listening layer that operates directly on raw audio. Modulate will demonstrate the platform at Customer Contact Week Las Vegas, June 22–25, 2026.
Our Analysis
This announcement is worth more attention than a typical API launch. Modulate is making a direct architectural argument: that speech-to-text transcription, the current foundation of most enterprise voice AI deployments, is structurally insufficient for real-time operational decisions. That argument is both technically credible and commercially well-timed.
The Transcript-First Problem Is Real
The dominant voice AI stack today follows a predictable pattern: audio in, transcript out, downstream NLP on top. That pipeline works well for search indexing, call summarization, and after-the-fact analytics. It fails badly for anything requiring real-time awareness. Hesitation, silence, emotional escalation, and the acoustic signatures associated with synthetic or manipulated audio do not survive transcription intact. By the time a transcript lands in a downstream model, those signals are gone.
Modulate’s ELM architecture aims to address this directly by running an ensemble of specialized models against raw audio rather than serialized text. The practical implication is that Velma can, in theory, flag a fraudulent interaction, a distressed customer, or a misbehaving AI agent while the conversation is still live. For contact centers, fraud operations, and compliance teams, that real-time window is the difference between prevention and after-the-fact remediation.
What This Means for ITDMs
The use case list in the announcement spans fraud detection, customer experience, AI agent oversight, trust and safety, compliance, and operational intelligence. That breadth is deliberate. Modulate is positioning Velma as horizontal infrastructure rather than a vertical point solution, and that positioning makes sense given where the market is heading.
Consider AI agent oversight specifically. As enterprises accelerate AI deployment across customer-facing channels, the question of how to supervise AI agents in real time becomes urgent. A transcript-based audit log catches problems only after they’ve already reached a customer. A continuous voice intelligence layer that flags policy violations or inaccurate claims during the conversation is a fundamentally different operational capability.
ECI Research’s 2025 AI Builder Summit survey found that 44% of enterprise AI leaders have only moderate confidence that AI agents can act autonomously without human intervention. That confidence gap is precisely the environment in which a real-time oversight layer like Velma finds its market. ITDMs evaluating AI deployment at scale should be asking whether their current observability and oversight stack extends to the voice channel, and in most cases today, it does not.
The compliance angle is equally compelling. Industries with vulnerable customer regulations, including financial services, insurance, and healthcare, face material liability when agents miss signs of distress, confusion, or coercion during live calls. Post-call review catches some of these failures. Real-time detection changes the outcome.
What This Means for Developers
Opening the API to any developer, not just enterprise contract customers, is the strategic move here. Modulate is seeding a developer ecosystem around a capability that has previously required direct vendor engagement. That matters because the most interesting enterprise deployments of voice intelligence will be built by teams integrating it into existing workflows: CRM platforms, fraud decisioning systems, AI agent orchestration layers, and contact center tooling.
From an implementation standpoint, the ensemble architecture is worth examining carefully. The claim that specialized models running in parallel outperform a single large model on latency and cost at production scale is plausible, and aligns with broader industry movement toward modular AI architectures. Developers building on the API should evaluate latency characteristics under realistic load, how structured outputs are formatted for downstream consumption, and what the explainability surface looks like for audit and compliance workflows. The press release references “explainable signals” as a design principle, which is the right framing for regulated industries but warrants verification in practice.
The expansion also touches on a theme that ECI Research has tracked consistently across AI/ML operations. According to ECI Research’s survey of 489 AI/ML practitioners and managers, 92% of organizations report that AI capabilities are now integrated into at least one stage of their software delivery lifecycle, a sharp increase from 71% in early 2024. Voice is one of the last major interaction channels without mature real-time AI instrumentation. Modulate is betting that developers will build that instrumentation on Velma rather than assembling it from generic components.
Competitive Positioning
Modulate enters a space that includes general-purpose voice AI providers, contact center platform vendors with embedded analytics, and specialized fraud detection vendors. Its differentiation rests on three claims: raw audio analysis rather than transcript dependency, an ensemble model architecture tuned for conversational dynamics, and a heritage in high-noise, high-stakes real-world audio (online gaming environments). That last point is unconventional as a credentialing narrative but actually relevant. Gaming voice environments are unscripted, emotionally variable, and adversarial in ways that enterprise contact center audio is not uniformly, but does resemble in important moments.
The risk is market education. Most enterprise buyers have built their voice analytics investments around transcript-based pipelines and will need a clear ROI narrative to justify architectural change. Modulate’s developer-first API expansion is likely a deliberate strategy to build proof points from the bottom up, letting integrated deployments generate the case studies that enterprise sales requires.
What’s Next
The Real-Time Voice Intelligence Market Is Still Early
The Velma API launch signals a market that is moving from proof of concept to production infrastructure, but it’s not there yet. Most enterprises still treat voice analytics as a retrospective reporting function. The transition to continuous, real-time voice intelligence will follow the same adoption curve as real-time application observability did earlier in the decade: slow initial uptake, then rapid normalization once a handful of high-visibility wins (a prevented fraud event, a caught compliance failure, a flagged AI agent hallucination) establish the category’s necessity.
ECI Research expects that the AI agent oversight use case will be the fastest-moving entry point. As agentic AI deployments scale across customer-facing channels, the demand for real-time behavioral monitoring of those agents will grow in lockstep. Voice channels, which carry more emotional and contextual signal than text, will be a priority surface for that monitoring. Velma’s positioning here is well-timed.
Developer Adoption Will Determine Enterprise Velocity
The decision to open the API broadly is the right move at this stage of the market. ECI Research’s analysis of AI/ML platform adoption consistently shows that enterprise deployment velocity tracks developer familiarity with underlying APIs. Modulate’s challenge over the next 12–18 months is conversion: turning developer experimentation into production integrations, and production integrations into enterprise contract conversations. The Customer Contact Week presence is a signal that the enterprise sales motion is running in parallel, but the developer ecosystem will be the longer-term moat. Teams building real-time customer experience, fraud, or compliance infrastructure should put Velma on their evaluation list now, before architectural decisions around voice pipelines are locked in.
Stay Ahead of Application Development Trends
Get weekly analyst insights, research notes, event coverage, and AppDevANGLE updates delivered directly to your inbox.
Subscribe for Weekly Insights
Join technology leaders, practitioners, and GTM teams following the trends shaping modern software delivery.
Looking for deeper research access?
Explore ECI Research reports, survey insights, and market analysis through the ECI Research Portal.
