How prathameshpalshikar/sravaani-onnx-live-test works
SraVaani Stage 2 is a local, CPU-based live microphone transcription runner for a pre-existing streaming ONNX INT8 export. The user selects a microphone and model, audio is captured, resampled and queued for a sherpa-onnx recognizer, and transcript updates and a session report are produced. Device listing and CPU inspection support setup and thread experiments. The main script excerpt is partial; links reflect the README and visible implementation, and uncertain internal wiring is omitted.
Main entry points: run_sravaani_onnx_live.py Live runner · system_info.py CPU information · results JSON report
Components 14
Live transcription
- Live runner
scripts/run_sravaani_onnx_live.py - Audio capture
scripts/run_sravaani_onnx_live.py - Resampling
scripts/run_sravaani_onnx_live.py - Bounded audio queue
scripts/run_sravaani_onnx_live.py - Inference worker
scripts/run_sravaani_onnx_live.py - Streaming recognizer
scripts/run_sravaani_onnx_live.py
Inputs and setup
- ONNX model files
- Device listing
scripts/run_sravaani_onnx_live.py - CPU information
scripts/system_info.py
Session results
- Live transcript
scripts/run_sravaani_onnx_live.py - Session metrics
scripts/run_sravaani_onnx_live.py - JSON report
results
Other components
- User
- Microphone
Connections 17
- User → Live runner: starts session
- Live runner → ONNX model files: validates files
- Live runner → Device listing: lists devices
- User → CPU information: requests CPU info
- Microphone → Audio capture: provides audio
- Live runner → Audio capture: opens stream
- Audio capture → Resampling: passes audio
- Live runner → Resampling: resamples audio
- Live runner → Bounded audio queue: enqueues blocks
- Bounded audio queue → Inference worker: delivers blocks
- Live runner → Inference worker: starts worker
- Inference worker → Streaming recognizer: decodes audio
- ONNX model files → Streaming recognizer: loads model
- Inference worker → Live transcript: sends updates
- Live runner → Session metrics: collects measurements
- Inference worker → Session metrics: sends metrics
- Live runner → JSON report: writes report