IBM announces ProsodyLM, AI-powered sports commentary featuring capabilities from computer vision and speech models for expressive narration at major tournaments.

IBM uses AI-powered commentary to pair computer vision and sophisticated speech models to provide almost real fiction narration at high-profile sporting events, including the US Open.

IBM researchers are training AI systems through its MIT-IBM Watson AI Lab to tailor tonal, pitch, and rhythmic adjustments in real-time, allowing commentary to appear more interactive and more human. Models examine the metadata of the game, crowd excitement, and player movements to produce dynamic scripts, which are then spoken through the new IBM ProsodyLM voice technology.

ProsodyLM is trained on 30,000 hours of speech information and is far more than just a text-to-speech system; it is a system that includes the prosody: intonation, rhythm, and emphasis, which are required to make a human utterance sound natural. BIB Research noted that the innovation allows the AI commentators to sound excited when the rally is exciting but precise in times of calm.

Wimbledon and the US Open 2023 were the first sporting events to use the technology of IBM Consulting and IBM Research. The AI, with its new expressiveness, is expected to provide wider coverage for courts that will lack human announcers, while not aiming to surpass professional commentators.

The system was demonstrated as part of the 2025 Behind the Scenes demo at IBM's 2025 US Open, where the system can be expanded beyond tennis to other sports, such as Formula 1. Newer models can even enable fans to tailor commentary styles, including low-key to extremely spicy.

With AI accuracy and expressiveness like a human being, IBM is establishing a new example of how AI can enhance the fan experience of watching live sport.