Speech analytics

How it works

How it works

Speech-to-text turns each recording into text, keeping the agent's and the customer's words apart. Then rules and AI models analyze the text and audio, and the results are saved as topics, tags, and scores. You search calls and build reports by them.

• Transcript – each speaker's text with timestamps
• Search – calls by words, phrases, and their combinations
• Topics – reasons for calling
• Speaking style – silence, interruptions, overtalk, who talks how much, pace
• Scoring – script, required phrases, and forbidden phrases
• Emotions – the customer's sentiment and how it changed during the call

Use

What it is used for

Practice

Getting value from it

The system finds patterns, but people still have to act on them. Start with 3-5 specific questions – for example, "why do customers call again?" or "do agents offer the upsell?" – set up topics for them, and review the results every week.

• Record the agent's and the customer's voices on separate tracks so the system does not mix up who said what
• Test recognition in English on your own calls, terms, and product names
• Start with a small set of topics and add new ones gradually
• Turn findings into agent coaching, script changes, and knowledge base articles

In CONTACT.WORK

How it works in CONTACT.WORK

Speech analytics works with recordings from the platform's telephony or your current PBX. Calls are recorded in stereo plus a separate MP3 track for each side; 1,000 hours take about 90 GB. The system transcribes calls, scores them against your QA scorecard, tags topics, and searches across all conversations. Where state law requires all-party consent, play a recording notice at the start of the call.

Quality scoring covers calls, chats, and email. Speech analytics scores conversations automatically, and reviewers add manual scores and comments where a person's judgment is needed. For every call, AI writes a short summary, and the summaries show up in reports.

You can run analytics in the CONTACT.WORK cloud or entirely on your own servers, so recordings stay inside your company. The platform itself is free. Speech recognition and the language model for analytics are paid separately or run on your own hardware. The server calculator sizes the hardware for analytics on your servers.