soundKIT

AI Playground for Sound

AI Playground for SoundAmbiq soundKIT™ is an open-source audio AI development kit (SDK) for training, evaluating, and deploying real-time, on-device audio models on Ambiq’s ultra-low-power SoCs. It supports speech enhancement, keyword spotting, and speaker verification for always-on edge AI applications, including smart home devices, wearables, smart glasses, industrial sensors, and medical devices.

soundKIT Highlights

Audio Edge AI

Develop always-on key AI audio features, including speech enhancement (denoising and dereverberation), keyword spotting, voice activity detection, and speaker/voice authentication, on edge devices.

Task-Level Modes

Use s a consistent CLI across the full audio AI workflow: dataset setup, model training, benchmarking, validation, deployment, and real-time inference demos on PC or Ambiq EVBs.

Flexible Iteration

Swap datasets, models, tasks, and training recipes using YAML configurations, CLI commands, or Python APIs, without changing the core workflow.

Power and Performance

Powered by heliaRT™ for production-grade deployment on Apollo SoCs, delivering up to 3× faster inference than TFLM with vector-accelerated AI kernels, and enhanced int16x8 quantization for high-fidelity audio and speech

Tasks & Capabilities

Speech Enhancement (SE)

On-device speech enhancement with denoising and dereverberation to improve speech clarity in noisy, real-world environments.

Keyword Spotting (KWS)

Fast, low-memory wake-word detection using lightweight models compatible with TensorFlow Lite and TensorFlow Lite Micro (TFLite/TFLM).

Voice Activity Detection (VAD)

Real-time, low-power detection of speech presence to reduce false triggers and extend battery life in always-on audio systems.

Speaker Verification (ID)

Secure, privacy-preserving on-device speaker verification for voice authentication—no cloud connectivity required.

Design Resources

Design Resources

Instant download

Additional documentation

Set favorites
Get update notifications
Use search filters

soundKIT

AI Playground for Sound

soundKIT Highlights

Audio Edge AI

Task-Level Modes

Flexible Iteration

Power and Performance

Tasks & Capabilities

Speech Enhancement (SE)

Keyword Spotting (KWS)

Voice Activity Detection (VAD)

Speaker Verification (ID)

Design Resources

Design Resources

Instant download

Additional documentation

Additional Resources

Video title