How the Voice Inbox Receives Messages
Every Voice Inbox has a unique recording endpoint — a URL that resolves to a minimal, single-purpose recording interface. When someone accesses that URL by scanning a QR code on a table card, tapping a link in an SMS, clicking a widget on your website, or following a link in an email, they land on a page with one function: record a voice message.
The interface renders in under 500 milliseconds, adapts to any screen size, and requests microphone permission only once. The person presses the record button, speaks, and releases. The audio is encrypted in transit over TLS 1.3 and arrives at VoiceZero's processing pipeline within milliseconds of the recording ending. No uploads to wait for, no confirmation steps, no "thank you — please complete this short survey" redirect that undermines the effort.
On the backend, the message enters a five-stage pipeline: transcription, sentiment scoring, theme tagging, urgency classification, and routing. All five stages complete in under three seconds. By the time the sender has put their phone down, the message is already organized, labeled, and visible in the inbox dashboard — with its full sentiment profile attached.
One inbox, all channels: QR codes, direct links, embedded widgets, and SMS links all feed the same inbox. Every message arrives with a channel tag so you know exactly where it came from — but the analysis, routing, and display is unified. No toggling between inboxes per channel.
Organizing Multiple Inboxes
A single VoiceZero account can contain as many inboxes as your organization needs. A hotel group might run one inbox per property, plus separate inboxes for housekeeping feedback, maintenance reports, and employee engagement. An HR department might maintain distinct inboxes for exit interviews, pulse checks, and confidential safety reporting — each with its own QR code, routing rules, and escalation contacts.
Each inbox has its own label, color code, assigned team members, and routing configuration. The main dashboard can display all inboxes simultaneously — sorted by recency, urgency, or volume — or you can drill into a single inbox for focused review. Analytics can be compared across inboxes to spot outlier locations or departments before problems compound.
The Recording Experience
The decision to keep the recording interface minimal was deliberate and data-driven. Every element added to a feedback interface — a logo, a form field, an optional name prompt, a star rating — reduces completion rates. VoiceZero's recorder presents exactly one thing: a prominent record button and a brief instruction line. That is it.
Microphone access is requested once. The visual indicator during recording is a simple animated waveform — enough to confirm the microphone is active, not enough to distract. Most voice messages run between 20 and 60 seconds. This is not a limitation — it is a feature. Voice feedback is naturally self-editing. People say what matters most and stop, unlike text surveys where respondents feel pressure to fill every field. You get more signal per response from voice than from any written format.
Recordings are capped at five minutes to prevent accidental over-length captures. If a sender stops mid-message, the partial recording is discarded locally — it is never sent unless the sender actively submits it. There is no partial-message leakage.
What the recorder collects
- Audio waveform of the voice message
- Recording timestamp (UTC)
- Inbox channel tag (QR / link / widget / SMS)
- Message duration in seconds
- Browser language setting (for transcription hint)
- Submission confirmation flag
What it never collects
- Name, email, or phone number
- IP address or geolocation data
- Device fingerprint or hardware identifiers
- Browser cookies or session tokens
- Account login or authentication data
- Any PII linkable to the sender
Multi-Stage Processing Pipeline
Once a voice message arrives, VoiceZero processes it through five sequential stages before it appears in your inbox. The entire pipeline runs in parallel across distributed infrastructure and completes in under three seconds for messages up to two minutes long.
Transcription
The audio is converted to text using a multilingual speech recognition model supporting 90+ languages. Language is auto-detected from audio features — no language selection required from the sender. The transcript is timestamped at the word level, enabling later analysis to correlate spoken words with vocal delivery patterns.
Sentiment Analysis
The AI Sentiment Engine processes both the transcript text and the raw audio in parallel. It outputs an overall valence score (positive, negative, neutral, or mixed), a primary emotion label from eight possible dimensions, and a confidence percentage for the classification. Voice-based analysis captures sarcasm, resignation, and mixed signals that text alone would miss.
Theme Tagging
The transcript is analyzed for topic categories relevant to your industry and inbox configuration. A hospitality inbox might tag messages with "cleanliness," "staff attitude," "noise," "amenities," or "value." An HR inbox might tag "management," "workload," "benefits," or "team culture." Tags can be customized per inbox or left to auto-detection from the base model.
Urgency Scoring
Every message receives a 0–100 urgency score based on both content signals (words indicating safety, immediacy, or distress) and vocal signals (elevated pitch, rapid speech, tense delivery). Messages scoring above configurable thresholds — typically 70 for standard escalation, 90 for immediate alert — are flagged automatically and trigger notification rules.
Routing and Escalation
Based on sentiment score, urgency score, detected themes, and inbox routing rules, the message is assigned to the appropriate team or individual. A maintenance complaint routes to facilities. A safety report routes to the safety officer with an immediate SMS alert. A positive message is tagged for sharing in the weekly digest. Smart escalation rules handle every routing decision without human triage.
📊 Inbox Analytics at a Glance
Every Voice Inbox includes a real-time analytics panel. These six metrics are always visible in your dashboard without needing to run a report or export data.
Anonymity Architecture
The Voice Inbox is built on a zero-knowledge anonymity model. This is not a privacy policy statement — it is a technical constraint. The system does not request, collect, or store any information that could be used to identify who sent a message. There is nothing to breach, subpoena, or accidentally expose.
No cookies are set during the recording session. No browser fingerprinting code runs on the recording page. The sender's IP address is used only to route the audio data to the nearest processing endpoint — it is discarded immediately after the connection is established and is never written to any database. The audio file is stored under a randomly generated UUID with no relationship to any session, device, or network identifier.
This architecture matters for the quality of your data, not just for compliance. People speak more honestly when they genuinely believe they cannot be identified. Anonymous voice feedback consistently surfaces concerns that named or semi-anonymous channels miss — because the social risk of speaking plainly has been removed entirely.
Encryption at every layer: Audio in transit is protected by TLS 1.3. Audio at rest is encrypted with AES-256 at the storage layer. Metadata (transcripts, sentiment scores, theme tags) is stored separately from audio and is also AES-256 encrypted. Decryption keys are managed via hardware security modules (HSMs) with no direct human access.
Export, API & Integrations
Voice messages do not have to live only inside VoiceZero. The platform exposes a full REST API that lets you pull message data — transcripts, sentiment scores, urgency ratings, theme tags, timestamps, and channel source — into any downstream system your team already uses.
- Tableau, Power BI, Looker
- Google Sheets export
- CSV / JSON download
- Scheduled data pushes
- Custom date range filtering
- Salesforce via webhook
- HubSpot contact tagging
- Zendesk ticket creation
- Freshdesk integration
- Custom CRM via REST API
- BambooHR workflow triggers
- Workday data push
- Slack channel alerts
- Microsoft Teams notifications
- Email digest to HR team
- Zapier (900+ app connections)
- Make (Integromat) flows
- Native webhooks (JSON)
- Trigger on urgency threshold
- Trigger on sentiment shift
Every API call is authenticated with a scoped API key. Read-only keys can be issued for BI tools without exposing write access. Webhook payloads are signed with HMAC-SHA256 so receiving systems can verify origin. Rate limits are generous — 1,000 requests per minute on Business plans, unlimited on Enterprise.
Deploy Your First Voice Inbox in 60 Seconds
Create an inbox, share the QR code or link, and receive your first analyzed voice message — all within one minute. No credit card required on the free plan.