← All case studies
Real-time applied AICase 09
NeMo ASR
Real-time audio streaming & ASR pipeline
A full-stack streaming pipeline connecting an Angular client to an NVIDIA NeMo speech-recognition server through Java Spring Boot WebSockets.
01 / 03Web and mobile clients stream audio through a Java middleware to NeMo and get text back.
Contribution
Full-stack prototype · real-time streaming · AI integration
Domain
Speech recognition · WebSockets · applied AI
Availability
Public demo and source repositories
The problem
Live audio requires low-latency, bidirectional communication. Request/response HTTP introduced delay and made it harder to coordinate chunked audio, middleware, and hardware-accelerated recognition while returning text continuously.
What we built
- WebSocket middleware: Built a full-duplex Spring Boot server for concurrent binary audio streaming and immediate transcription messages.
- Hardware-accelerated AI bridge: Connected enterprise Java middleware to a Python-based NVIDIA NeMo ASR service.
- Reactive Angular client: Captured microphone input and streamed chunked packets without blocking the user interface.
- Cross-stack event coordination: Standardized message structures across TypeScript, Java, and Python to preserve ordering and consistency.