On Feb 24, 2026, IBM and Deepgram announced a collaboration to bring Deepgram's speech-to-text and text-to-speech technology into IBM's watsonx Orchestrate platform. The integration is aimed at enterprise AI teams that want voice interfaces, transcription, and real-time captioning inside the same orchestration stack they already use for digital agents and workflow automation.

IBM says it will embed Deepgram's capabilities into watsonx Orchestrate to support enterprise-grade transcription and real-time captioning. Deepgram becomes IBM's first voice partner under this arrangement, which is a meaningful signal about how IBM wants to extend watsonx. Instead of building every modality alone, IBM is combining its orchestration layer with a specialist provider where speech quality, latency, and reliability matter most.

The announcement also shows where enterprise AI demand is shifting. Both companies frame voice as a default interface for practical systems, not just a convenience feature. That includes digital agents that accept spoken instructions, internal assistants that summarize or route conversations, and customer-facing workflows where low-latency recognition and natural audio output make automation usable in live interactions. IBM is positioning watsonx Orchestrate as the layer that manages those flows, while Deepgram supplies the speech stack underneath.

Deepgram's CEO said enterprise deployments need a real-time platform that is accurate, low latency, and reliable at scale. IBM's partnership team described the integration as a way to modernize operations while preserving customer choice inside an open ecosystem. Those statements matter because they frame this as production infrastructure for large organizations rather than a demo for isolated chatbots.

The practical implication is that voice AI is moving deeper into mainstream enterprise software. Instead of treating speech recognition and text-to-speech as separate add-ons, vendors are folding them into orchestration platforms where models, agents, and business processes are already managed. If IBM executes well, the Deepgram partnership could make voice a native part of enterprise agent deployment instead of a specialized side project.

#deepgram

IBM and Deepgram bring speech-to-text and text-to-speech into watsonx Orchestrate