Spaces:
Sleeping
Sleeping
Commit History
-
Add Trends tab with analytics charts and Coming Soon sections
- Risk Score Over Time: line chart with green/yellow/red zones - Fraud Type Distribution: stacked bar chart by day - Voice Reuse Patterns: top 10 voices bar chart with color coding - KPI summary: total tests, avg risk, high risk %, recurring voices - Coming Soon: Batch Analysis and Webhook/API Integration with impact Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>5b529ce -
Hard fraud indicators force MEDIUM risk minimum
- 2+ speakers detected: minimum 45 (MEDIUM) - Whispers detected: minimum 40 (MEDIUM) - Wake word detected: minimum 35 (MEDIUM) These are clear fraud signals that should never result in LOW risk. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>1d50892 -
Wake word detection always triggers MEDIUM risk minimum (35)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>2c0695b -
Fix PDF download: convert bytearray to bytes for Streamlit
fpdf2's output() returns bytearray which Streamlit's download_button does not accept. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>4d50a3e -
Add Test History with View Report in Database tab
- Add AnalysisResult.from_dict() to reconstruct results from stored JSON - Add Test History section in Database tab listing all past analyses - View Report button loads full report inline with back navigation - Results now persist across Streamlit reruns via session_state Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>91ee436 -
Fix bytearray crash in voiceprint matching
SQLAlchemy LargeBinary returns bytearray, but np.frombuffer requires bytes. Wrap with bytes() before calling np.frombuffer. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>231585e -
Add Tier 2: Comparison View tab and Alert Rules Engine
- New "Compare" tab: side-by-side test comparison with risk scores, detection flags, speaker summary, shared voice detection, and grouped bar chart breakdown - Alert Rules Engine in sidebar: configurable thresholds for risk score, synthetic voice %, max pause duration, speaker count, and recurring voice appearances - Triggered alerts shown in results when rules are violated - DB methods: get_all_tests() and get_test_results() for comparison data Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>f565754
-
Add Tier 1 improvements: Risk Score, KPIs, PDF, Demo Samples
Feature 1 - Consolidated Risk Score (0-100): Weighted score from synthetic, playback, reading, whispers, pauses, wake words. Color-coded banner (green/yellow/red). Feature 2 - Demo Audio Samples: 3 pre-generated TTS demos (clean, reading fraud, coaching fraud) built at Docker build time via gTTS. "Try a Demo" selector in UI. Feature 3 - Dashboard KPIs: Tests Analyzed, Fraud Rate %, Unique Voices, Alerts Today displayed at top of Analyzer tab. Feature 4 - PDF Report Export: Professional downloadable PDF with risk score, speaker summary, detection flags table, alert details. Uses fpdf2. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>3bafd90 -
Fix audio playback + persist logs across refreshes
Audio clips: Convert from float32 WAV to PCM int16 WAV for browser compatibility. Float32 WAV causes playback to cut short or fail in many browsers. Logs: Save session history to JSON file on disk (DATA_DIR/logs/) so logs survive page refreshes and container restarts. Load from disk on init, clear button also wipes the file. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>89a987c -
Fix permission denied: store models/data outside /app mount
HF Spaces mounts repo content over /app/ at runtime, overwriting Docker build-time permissions. Move model downloads to /home/user/models/ and data storage to /home/user/data/ which persist from build. All source files now use MODEL_DIR and DATA_DIR env vars for path resolution. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>9109931 -
Add technical Logs tab with live terminal and session history
Adds a 4th tab "Logs" showing detailed pipeline logs for tech lead review. Each analysis step logs the library, model, parameters, and results (e.g. SpeechBrain VAD segments, ECAPA-TDNN embeddings, Whisper transcription). Logs are grouped by analysis session with collapsible history and a live terminal view. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>d4458bf
-
Add SOP Audio Analyzer app files
- Full Streamlit app with audio analysis - SpeechBrain VAD, diarization, voiceprint - Fraud detection modules - Database modelsebba35f -
Duplicate from streamlit/streamlit-template-space
Verified signatureCo-authored-by: Franck Abgrall <FranckAbgrall@users.noreply.huggingface.co>c5038fa
