Web & apps · 2023
Randstad Voice Pipeline
Real-time voice pipeline for Randstad: speech recognition and neural text-to-speech, streamed over WebSockets.
A client project with DIRIS for Randstad: a GPU voice pipeline that listens and speaks in real time.
- Speech-to-text with Whisper or NVIDIA NeMo, switchable
- Neural text-to-speech with Coqui TTS on CUDA, accelerated with DeepSpeed
- Audio clean-up with noise reduction before transcription
- Real-time streaming over WebSockets