Phase 1 · Local model quality test
Bingo Audio Test v0.7.5.2
Whisper runs locally. v0.7.5.2 adds mixed-precision model loading for Base and Small: a compressed encoder with a full-precision decoder to bypass the iPhone/WASM quantized-decoder failure.
1. Load the local speech model
WASM remains the safest iPhone backend. Select a model, load it, upload the same saved recording, and rerun the audit. Only one model is retained at a time to reduce memory pressure.
Model not loaded.
2. Record a short benchmark clip
Use this for quick individual-call checks. Voice enhanced requests automatic gain and speech cleanup.
3. Transcribe and evaluate the short clip
Enter each intended bingo call once. Repetition is evaluated separately.
Raw Whisper transcript
Parsed call events
Elks repeat status
4. Automated 75-call audit
Use the earlier sequential or randomized audit audio on the laptop. The tracker now analyzes overlapping audio segments and aligns detected calls by order, so variable voice pacing and start delays do not break the audit.
Uploaded recording model test
No recording selected. The same selected file can be rerun after loading another model.
Laptop-speaker room test
Tap Start room recording, then play the earlier sequential or randomized audit audio on the laptop. No beeps are required. Stop the phone recording after the caller says Done. The 12-minute safety limit is only a backup.
0/75 calls · 0%
Current call transcript
Missed calls
Provisional-only calls
Unmatched detected calls
Results by letter
5. Parser-only test
Paste a Whisper transcript here to test the bingo parser without recording again.