Introduction
What VoiceLens does, who it is for, and how a call travels from your phone system to a graded scorecard.
VoiceLens is a quality layer for voice calls. You send it recordings (or let your telephony platform send them), it transcribes and analyses each one, and it returns a scorecard: a health score from 0 to 100, a list of detected faults with timestamps, per-turn latency statistics, the details the agent was supposed to capture, and the caller's sentiment over time.
It does not matter whether an AI voice agent or a human agent was on the line. The same rubric is applied to both, so you can compare a bot against a team, or one prompt version against another, on identical criteria.
Who uses it
- Teams running AI voice agents who need to know where the bot repeats itself, goes quiet, ignores a correction or answers too slowly, across every call rather than a sample.
- Contact-center QA and coaching teams who want every human agent call graded with evidence pinned to the transcript.
- Compliance teams who must verify that required phrases were said and forbidden ones were not.
- Agencies and BPOs who run a workspace per client and need white-label reporting.
How a call flows through VoiceLens
A call arrives
Upload a recording in the dashboard,
POSTit to the ingest API with a channel token, or point your Twilio recording callback at VoiceLens. Calls belong to a channel, which represents a phone line, a bot, or a team.It is transcribed
Audio is transcribed with speaker diarization, word-level confidence and (for English) per-turn sentiment. Line quality is measured acoustically at the same time.
It is graded against a rubric
The channel's rubric template defines the slots to capture, the questions to ask, the phrases that are required or forbidden, and the latency thresholds. Detectors then look for the thirteen fault codes.
You get a scorecard
A health score, faults with evidence, latency statistics, slot and question coverage, and sentiment appear in the dashboard, and are pushed to your systems through webhooks and the read API.
Key terms
| Term | Meaning |
|---|---|
| Workspace | An isolated container for calls, channels, templates, members and keys. Agencies use one per client. |
| Channel | A source of calls: a phone number, a bot, a queue. Each channel has ingest tokens and a rubric. |
| Rubric template | The definition of a good call for a given call type: slots, questions, compliance phrases, thresholds. |
| Conversation | One analysed call. The dashboard calls them calls; the API calls them conversations. |
| Fault | A detected problem on a call, with a code, a severity (danger, warning or info) and evidence. |
| Health score | The 0 to 100 grade computed from faults, slot capture and question coverage. |
| Golden set | Calls you have graded by hand, used to calibrate a rubric. |