Webinar: A new era for audio AI data

On the agenda for this Webinar:

• Audio that moves - finally
  Synthetic audio has traditionally been static, unable to represent the fluid nature of sound in real life. People move. Devices move. Conversations shift direction. Acoustic conditions change continuously. In this webinar, Treble will introduce world-first dynamic scene modeling, enabling simulation of moving sound sources, moving listeners, and time-varying spatial relationships. This unlocks more realistic training data and enables virtual prototyping of audio devices operating in dynamic environments - a fundamental expansion of what synthetic audio can represent.
• Curated audio data, directly accessible
  Treble is evolving from a simulation SDK into a full Audio AI Data Platform. We will show how teams can directly access large-scale, curated audio and acoustics datasets and work with them alongside simulation and data generation. The session will demonstrate how data can be processed and tailored to specific use cases, with metadata treated as a first-class object that can be customized and enriched. Attendees will see how this enables faster iteration, better collaboration, and training datasets designed for real-world conditions. 
• Research spotlight: why better data wins
  Treble will present new research showing measurable improvements in downstream ML performance when training on Treble data, including gains in monaural and multi-channel speech enhancement, speech recognition, direction-of-arrival & distance estimation, and blind room estimation. We will discuss why, as model architectures converge, data quality and diversity become decisive for audio AI performance.
• New data-centric ML workflows
  High-quality data only matters if teams can use it efficiently. We will introduce new data-centric workflows for handling and processing large audio datasets at scale, improving collaboration between data, simulation, and ML teams, and integrating Treble seamlessly with existing training infrastructure. Attendees will see how these workflows reduce friction and improve reproducibility in real ML pipelines.
• Own-voice modeling and enhanced device handling
  We will showcase new capabilities for modeling challenging device-level audio scenarios, including devices with microphones and speakers and complex near-field and own-voice interactions. The session will demonstrate how detailed geometry and acoustic configurations can be handled efficiently, improving both training data quality and realism in virtual device prototyping.
•  Automated scene generation
  We will introduce new tools for automated virtual audio scene generation, enabling teams to construct nuanced audio scenes at scale by controlling SNR, speech overlap, turn-taking, and environmental complexity, while incorporating real speech data where needed. We will show how this integrates with Treble’s 3D world capabilities, including room databases, programmatic environment generation, and CAD and OpenUSD-based workflows.