Bringing Ode Poetry to life with MAI’s audio models

William Sieghart, founder of National Poetry Day and author of *Poetry Pharmacy*, collaborated with Microsoft AI to create Ode Poetry, a digital platform that replicates his one-on-one poetry consultations. Users describe their emotions, and the system recommends a poem read aloud by a recording of Sieghart’s voice, maintaining the warmth and personalization of his in-person sessions. The project aims to make poetry accessible to a broader audience while preserving the human touch of Sieghart’s approach.

Ode Poetry leverages MAI’s MAI-Transcribe model to process spoken input in real time, handling diverse languages, accents, and dialects with a Word Error Rate of 4.86% on the FLEURS benchmark. This ensures natural, unstructured conversations without requiring users to adapt their speech. The model’s performance in production environments demonstrates its reliability for interactive applications like Ode Poetry, where understanding nuanced emotional cues is critical.

To recreate Sieghart’s distinctive vocal style, Ode Poetry uses MAI-Voice, which generates high-fidelity audio from minimal samples. The model was trained on a multilingual data pipeline and studio recordings, enabling it to faithfully reproduce Sieghart’s intonation and pauses. This capability was refined through iterative testing to ensure the digital experience closely mirrors his live consultations, enhancing authenticity and emotional resonance.

The experience was developed with responsible AI practices, including safeguards against misuse such as inappropriate poem recommendations or simulated diagnoses. Microsoft AI and Sieghart’s team refined prompts, recommendation logic, and speech output through extensive testing across emotional scenarios. Ode Poetry launched today, offering users a new way to engage with poetry while advancing Microsoft AI’s mission to create impactful, human-centered technology.