Voice Analytics
ProBeya Voice Analytics gives OpEx coordinators and team leads a dedicated dashboard to understand how voice commands are being adopted across the organization. By surfacing usage trends, error patterns, and per-user adoption metrics, the feature enables targeted training and vocabulary improvements that boost hands-free productivity in cleanroom and production environments.
Overview
In pharma and manufacturing, operators often work in environments where touching a screen is impractical -- gowning zones, cleanrooms, or active production lines. ProBeya's voice interface lets operators create actions, update KPIs, and log issues by speaking. But voice adoption requires active monitoring: operators who struggle with recognition may silently abandon voice and return to manual entry.
Voice Analytics solves this by aggregating transcription data into actionable charts and tables, letting administrators identify recognition gaps, track adoption progress, and measure the ROI of custom vocabulary investments.
Getting Started
- Enable voice features -- Ensure your organization has the voice module activated and the
VOICE_WHISPER_ENABLEDenvironment variable is configured. - Navigate to analytics -- Go to
/{workspaceSlug}/voice/analyticsfrom the workspace sidebar. - Review the dashboard -- The page loads five analytics panels automatically, drawing from the last 30-90 days of voice session data.
- Identify gaps -- Use the Error Patterns panel to find utterances that consistently fail recognition.
- Improve vocabulary -- Add custom vocabulary terms for frequently misrecognized phrases.
Key Concepts
- Voice Session -- A continuous period of voice interaction. Each session contains one or more utterances and records device type, locale, and cleanroom mode.
- Utterance -- A single spoken command within a session, including the transcribed text, detected intent, execution status, and confidence score.
- Intent -- The recognized action category (e.g.,
create_action,update_kpi,assign_task). Unrecognized utterances have a null intent. - Confidence Score -- A 0-1 value from the Whisper transcription engine indicating how certain the system is about the transcription accuracy.
- Custom Vocabulary -- Organization-specific terms (drug names, equipment IDs, project names) that improve speech recognition accuracy.
How It Works
Usage Trends Chart
Displays daily voice command counts over a configurable period (default: 30 days). The chart shows both session count (how many voice interactions started) and command count (total utterances). A rising trend indicates growing adoption; sudden drops may signal infrastructure issues or user frustration.
Success Rate Breakdown
A donut chart showing the proportion of successful, failed, and unrecognized commands over the last 30 days. Successful commands had their intent recognized and were executed. Failed commands were recognized but could not be executed. Unrecognized commands had no detectable intent.
Top Commands
A ranked bar chart of the 10 most frequently used voice intents. This reveals which voice commands deliver the most value and helps prioritize vocabulary expansion for underserved command categories.
User Adoption Table
Per-user metrics over the last 90 days showing:
- Total command count per user
- Individual success rate percentage
- Average confidence score
- Most recent session timestamp
This identifies users who need training (low success rate) and power users who can champion voice adoption.
Error Patterns
Lists the 20 most frequently failed utterances grouped by text (case-insensitive). Each entry shows the attempted intent, failure count, and average confidence score. Repeated failures on the same phrase indicate a systematic recognition gap that a custom vocabulary term could fix.
Configuration
| Setting | Description | Default |
|---|---|---|
VOICE_WHISPER_ENABLED | Enables server-side Whisper transcription | false |
| Analytics date range | Usage stats query window | 30 days |
| User adoption window | Per-user metrics lookback | 90 days |
| Error patterns limit | Maximum failed utterances shown | 20 |
| Top commands limit | Maximum intents in the ranking | 10 |
Vocabulary can be managed from the cleanroom settings page or via CSV import. The seedVocabulary procedure auto-populates terms from existing KPI names, project names, and team member names.
Permissions
| Role | Access |
|---|---|
| Organization Owner | Full access to analytics and vocabulary management |
| Tenant Admin / Site Admin | Full access to analytics and vocabulary management |
| Team Lead / OpEx Coordinator | View analytics dashboard |
| Operator | No access to analytics (can use voice commands only) |
All voice analytics queries are organization-scoped via orgProcedure. Voice data may contain sensitive spoken commands about production data and never leaks across tenants.
Tips & Best Practices
Use the Seed Vocabulary feature to automatically import KPI names, project names, and team member names from your existing ProBeya data. This gives immediate recognition improvement with zero manual effort.
Review the Error Patterns panel at least weekly during initial rollout. Each recurring failed phrase is an opportunity to add a vocabulary term that will improve success rates for all operators.
Use the User Adoption table to identify operators with low success rates. Pair them with a power user for a brief coaching session -- often the issue is speaking pace or microphone distance, not the system itself.
If your organization has existing terminology databases (drug names, equipment catalogs), export them as CSV and use the bulk import feature. The import uses ON CONFLICT DO NOTHING, so duplicates are safely skipped rather than causing errors.
If your workforce speaks multiple languages, compare success rates by locale. Some languages may need additional vocabulary tuning. The session filter supports locale-based queries to isolate language-specific recognition issues.
Troubleshooting
| Issue | Cause | Solution |
|---|---|---|
| "Whisper transcription is not enabled" | The VOICE_WHISPER_ENABLED environment variable is not set | Contact your administrator to enable the Whisper API integration |
| Dashboard shows no data | No voice sessions have been recorded yet | Ensure operators are using the voice interface; check that sessions are being created |
| High unrecognized rate | Operators using phrases the system does not map to intents | Review Error Patterns and add custom vocabulary terms for common phrases |
| Low confidence scores | Background noise or microphone quality issues | Ensure devices have adequate microphones; consider dedicated headsets for noisy environments |
| User adoption stalled | Operators reverting to touchscreen after failures | Address the top 5 error patterns first, then re-train affected operators |
Related Features
- Cleanroom Management -- GxP-compliant cleanroom mode with audit trail for voice commands
- AI Intelligence -- AI-powered insights and natural language queries
- KPI Boards -- The KPI definitions that operators reference in voice commands
- PWA & Offline -- Offline-capable progressive web app for shopfloor devices