Yes. Treble lets you define microphone positions on your device and compare how different configurations affect the captured sound.
Treble for Smart Glasses
Build smart glasses for the real world
Evaluate your complete audio and voice system, generate the data your models need, and develop differentiated features. Connect system testing, AI/ML development and hardware design to build experiences that work across real-world conditions.
The Challenge
Smart glasses have to work everywhere. Audio performance changes depending on the wearer, glasses fit, head movement, room, noise and talker position. Capturing the wearer’s own voice adds another challenge where the mouth is close to the microphones, while the head, torso and frame shape what each microphone receives.
Physical prototyping is slow and resource intensive, and recording campaigns cover only a fraction of these combinations. The gaps lead to failed voice commands, poor calls and algorithm edge cases when the product reaches users. The challenge is to understand performance in the conditions you haven’t tested yet.
How Treble Works
In Treble you can place smart glasses designs and wearer in realistic acoustic environments, and simulate what each microphone would capture. Systematically vary the wearer, fit, room, noise, talker position and motion to explore the conditions that matter to your product.
Use the resulting device-specific recordings to evaluate system performance, reproduce failure cases and generate data for AI/ML development. Compare hardware and algorithm changes under repeatable conditions, complementing physical testing throughout development.
Configure
Define your smart glasses and wearer, microphone configuration, acoustic environment and sound sources. Set up the fit, positions, orientations and operating conditions you want to explore.
Simulate
Use high fidelity acoustic simulation to model how sound interacts with the device, wearer and environment. Generate acoustic responses that capture what reaches each microphone.
Render
Combine simulated responses with speech, noise and other audio to create device-specific multichannel recordings, with clean references and structured metadata.
Evaluate, Train & Iterate
Use the recordings to evaluate system performance, train and improve algorithms, and identify failure cases. Compare changes under repeatable conditions and iterate on your product.
From microphone placement to real world speech performance.
Built for smart glasses audio development
Connect hardware design, system testing and AI/ML development. Understand what your device captures, generate the data your models need, and evaluate audio and voice experiences across a broader range of real-world conditions.
Explore microphone designs before building them.
Microphone Array Design
Compare microphone positions and frame geometries to understand how design choices affect the sound reaching your array. Evaluate configurations with your algorithms under the same acoustic conditions, without building a new prototype for every change.
Understand how your glasses capture the wearer’s voice.
Wearer Voice Capture
Model sound travelling from the wearer’s mouth to each microphone across different head and torso geometries, glasses fits and orientations. Account for near-field interactions, diffraction and acoustic shadowing when developing and evaluating voice capture.
Train with audio that reflects your device.
Device Specific Training Data
Create labelled multichannel recordings tailored to your device, wearer and use case. Generate difficult and underrepresented conditions systematically, with clean references and structured metadata for training, augmentation and evaluation.
Find failure cases before deployment.
Real-World Acoustic Complexity
Evaluate your complete audio and voice system across competing speech, background noise, reflections and movement. Reproduce challenging conditions on demand and compare performance as hardware, algorithms and software evolve.
Purpose built for smart glasses workflows
Inside Treble
Better Design. Better Listening.
Lower Word Error Rate
0%
Lower source distance estimation error
0%
Lower source localization error
0%
Resources and validations
Improving multichannel speech enhancement through accurate room-acoustic simulations
Training on the high-fidelity dataset results in an up to 38% relative reduction in median word error rate compared to the lower-fidelity alternatives. These results show that augmentation with high-fidelity room-acoustic simulations directly translates into improved multichannel speech enhancement performance.
Synthesis of Room Acoustics for Speaker Distance Estimation
Explore how simulated acoustic data can support models that estimate a speaker’s distance from a microphone. This study presents a benchmark for generating room acoustics and evaluating their value for speaker distance estimation.
Room-Acoustic Simulations as an Alternative to Measurements for Audio-Algorithm Evaluation
Evaluate audio algorithms with simulated acoustic data. See how wave-based simulations closely match measured results.
Frequently Asked Questions
Yes. You can include head and device geometries and model the acoustic path from the wearer’s mouth to the microphones, including near-field interactions and head shadowing.
Yes. You can configure surrounding sound sources and vary their positions, orientations, and acoustic environments to create repeatable test scenarios.
Device specific audio can support training and evaluation for beamforming, speech enhancement, source separation, and speech recognition.
