Pillar 3 — Understands
Automatic call analysis
Every inbound and outbound call turns into structured data automatically: speech recognition across several engines, speaker separation, voice identification, 49 system metrics, your own custom metrics, and topics and tags applied by meaning. This is the core of the platform — it lets you analyze 100% of your conversations instead of the usual 2–3% sample, without listening to anything by hand.
- Agent 0:02
Meridian Clinic, good afternoon. How can I help?
- Client 0:06
Hi, I'd like to book an appointment with a GP this week.
- Agent 0:11
Of course. There's a slot Thursday at 3:30 pm — does that work?
- Client 0:17
Yes, Thursday works. Please book it for me.
- Agent 0:21
Booked. I'll send a confirmation by SMS. Have a great day!
Key facts
- Coverage
- Automatic analysis of every call in full instead of a manual 2–3% sample — QA and analytics across 100% of conversations.
- Recognition
- Several STT engines with an accuracy-vs-cost choice, a repeat cache and VAD silence compression to save on recognition.
- Speakers
- Automatic diarization, a registry of speakers and voice prints, a voice map, and speaker-to-recording linking.
- 49 system metrics
- Conversation dynamics, prosody, sentiment, content and compliance — computed automatically on every recording.
- Custom metrics and topics
- 12+ custom metric types filled by AI, plus automatic semantic tagging of topics and tags.
Transcription tuned for accuracy and budget
The conversation becomes accurate text with the engine you choose — you balance quality and cost, and the platform strips out wasted spend.
- Several recognition engines to choose from: sharper where accuracy matters, cheaper where price matters.
- A cross-user repeat cache: the same audio is recognized instantly and for free.
- VAD silence compression before recognition saves roughly a fifth on long calls.
Diarization and voice identification
The platform doesn't just transcribe — it splits the conversation by role and recognizes who is speaking, call after call.
- Automatic diarization lays out utterances by speaker within each recording.
- A registry of speakers and voice prints recognizes returning participants and links them to recordings.
- The voice map is a clear 2D visualization that shows voice clusters and proximity.
49 system metrics by category
A set of metrics is computed on every recording automatically — what only an experienced supervisor used to hear is now measurable for every call.
- Conversation dynamics: silence share, longest silence, interruptions, overtalk, longest monologue, response latency.
- Prosody and emotion: speech rate, filler words, loudness, pitch variability, overall sentiment and its trend.
- Content and compliance: AI summary, topics, keywords, entities, action items, PII detection and profanity detection.
Custom metrics and semantic topics
The system set isn't enough? Describe what matters to you, and it gets filled on every call automatically.
- 12+ custom metric types: text, number, percent, yes/no, date, phone, URL, lists and more.
- AI extraction fills your metrics straight from the transcript instead of by hand.
- Automatic semantic tagging of topics and tags — up to a hierarchical call taxonomy.
Product
Automatic call analysis
- 1
The call enters the loop
The recording arrives from telephony, an integration or an upload — nothing extra to click.
- 2
Recognition and diarization
The chosen engine transcribes the speech, and diarization splits the conversation by speaker and recognizes voices.
- 3
Metrics, topics and tags
49 system metrics and your own metrics are computed on the recording, and topics and tags are applied by meaning.
- 4
Ready for analytics
The scored data syncs into analytics, dashboards and semantic search across the entire archive.
FAQ
Frequently asked questions
Why is this better than listening to a sample of calls?
Manual review usually covers 2–3% of conversations — the rest stays a blind spot. Here every call is transcribed and scored with metrics and topics automatically, so you control 100% of conversations and catch problems a sample would never reach.
Which recognition engines are available and how do I choose?
Several system engines are available, plus the option to plug in your own key. You balance accuracy and cost for the task; the repeat cache and VAD silence compression further cut recognition spend.
What are diarization and voice identification?
Diarization automatically splits the conversation by speaker. A registry of speakers and voice prints recognizes returning participants from call to call and links them to recordings, and the voice map shows this clearly in 2D.
Which metrics are computed automatically?
49 system metrics by category: conversation dynamics (silence, interruptions, overtalk, monologues, response latency), prosody (rate, filler words, loudness, pitch), sentiment and emotion, content (summary, topics, entities, action items) and compliance (PII, profanity).
Can I add my own metrics?
Yes. 12+ custom metric types are supported — text, number, percent, yes/no, date, phone, URL, lists and more. AI extraction fills them straight from the transcript, and topics and tags are applied by meaning automatically.
Related capabilities
Ready to start?
Turn every conversation into data, knowledge and action
Start by analyzing your conversations — no risk, no bots required. The platform turns your archive into data and a knowledge base, and voice agents plug in when you're ready.
Free plan, no card required.