AI Expertise & Services
When an off-the-shelf model isn't enough.
Cobalt's team works across acoustic modeling, language modeling, NLU, text-to-speech and speaker biometrics. We adapt existing products, or develop new ones, for the environments where general-purpose speech APIs fall short.
Where we help
Problems we take on.
- Domain-specific terminology
- Noisy acoustic environments
- Alternate microphone configurations
- Accented speech
- Children's speech
- Dysarthric speech
- Specialized applications
- Embedded and edge environments
Technical scope
What we work on.
- Automatic speech recognition (ASR) — hybrid and end-to-end
- Acoustic models and language models
- Natural language understanding (NLU)
- Text-to-speech (TTS)
- Voice biometrics and speaker diarization
- Deep neural networks
- Embedded deployment on constrained hardware
How engagements work
A typical path.
01
Scoping
We start with the data, environment and constraints — not the model.
02
Adaptation or R&D
Adapt an existing Cobalt product, or take on a Cobalt Labs research engagement.
03
Deployment
Ship the result into your infrastructure with the deployment model you require.