March 2025 Summaries

Deepgram and Opus Research’s 2025 survey of 400 North American business leaders, primarily from large enterprises, portrays voice AI as an increasingly foundational business technology, with 97% of respondents already using some voice-related tools and 67% considering them central to long-term strategy. The report identifies strong momentum toward AI-powered voice agents as replacements for traditional IVR systems, which remain widely used for customer service and task automation but achieve high satisfaction from only 21% of organizations. More than 84% of respondents expect to increase voice-technology budgets within a year, while companies developing voice agents largely plan to deploy them in production over the same period. Key adoption factors include customization, integration with existing systems, performance quality, compliance, and accessibility, with organizations applying voice agents to tasks such as order processing, FAQ responses, sales support, insurance quoting, and medical appointment scheduling. The findings suggest that businesses view natural, context-aware voice interactions as a way to improve efficiency, customer experience, accessibility, and regulatory compliance, making voice AI agents an anticipated mainstream enterprise capability rather than an emerging novelty.

Deepgram’s Genesys Transcription Connector integrates Deepgram’s real-time automatic speech recognition technology directly into the Genesys customer experience platform, allowing organizations to process voice interactions without third-party transcription tools or manual workflows. Powered by the Nova-3 streaming model, the connector is presented as offering low word error rates of 5.3% for clean audio and 6.8% for noisy audio, supporting accurate transcription in contact-center and other voice-intensive settings. Real-time voice data can improve automation for intent detection, sentiment analysis, routing, virtual assistants, and routine customer inquiries, while enabling agents to focus on more complex cases. Designed for native deployment and scalability, the integration aims to help organizations of varying sizes process high call volumes efficiently, improve response times, and support more personalized customer interactions. Activation requires coordination with a Genesys team, use of Deepgram’s setup documentation, and support from both companies during rollout.

Deepgram has launched Nova-3 Medical, a healthcare-focused speech-to-text model for developers building clinical voice AI applications, emphasizing accuracy, real-time performance, customization, and HIPAA-compliant security. Designed to handle noisy settings such as hospitals and telemedicine environments, it filters ambient noise while recognizing medications, diagnoses, procedures, and other specialized terminology. Its Keyterm Prompting feature allows users to add up to 100 custom terms without retraining, while deployment options include managed cloud, on-premises, and virtual private cloud environments. Deepgram reports that the model achieved a 3.44% median word error rate, a 6.79% keyword error rate, and a 93.99% keyword recall rate in its medical-audio benchmarks, outperforming named competing and previous-generation models according to its evaluation. The company positions the model for uses including clinical documentation, scheduling, prescription-related commands, telemedicine, and EHR integration, with pre-recorded transcription pricing beginning at $0.0043 per minute.