Speech analytics is software that transcribes phone calls and analyzes what was said and how it was said. Contact centers use it to find why customers call, check compliance, score quality, and coach agents.
It is the voice part of interaction analytics. Post-call speech analytics works on recordings, often across every call. Real-time speech analytics works during the call and can prompt the agent while the customer is still on the line.
How speech analytics works
Capture. The system takes the call audio, often with the agent and the customer on separate channels. That makes it easier to tell who said what.
Transcription. Speech recognition turns the audio into text. Accents, line quality, crosstalk, and product names all affect accuracy, which is usually measured as word error rate.
Analysis. The system looks for words and phrases, sorts calls into topics and contact reasons, and estimates sentiment. It also reads features of the audio itself: silence, people talking over each other, hold time, and the share of time each side speaks.
Action. Results feed dashboards, quality scorecards, alerts, and tags on the customer record.
What speech analytics is used for
Contact reasons. Categorizing every call shows what drives volume, often more reliably than the codes agents pick at the end of a call.
Compliance. Checking that required statements were made, such as a disclosure before a payment plan or the terms of a financial product.
Quality and coaching. Scoring every call against a checklist, then reviewing the calls where an agent struggled.
Customer risk. Flagging calls that mention cancellation, a complaint, or a sign that the customer may need extra support (see vulnerable customer detection).
Automation candidates. Finding frequent, simple requests that an AI agent could handle. See call center automation.
Speech analytics and the EU AI Act
Emotion detection has its own rules under EU law. The AI Act bans AI systems that infer the emotions of people in the workplace, except for medical or safety reasons (Article 5(1)(f)). The ban has applied since February 2, 2025.
The Commission's guidelines on prohibited AI practices give call center examples. A call center that uses voice recognition to track its employees' emotions, such as anger, falls under the ban. Tracking customers' emotions, such as anger or impatience, is not prohibited by Article 5(1)(f).
The guidelines read "workplace" broadly, whatever the person's status: employee, contractor, or trainee. Where emotion recognition is used on customers, Article 50(3) requires the deployer to inform the people exposed to it.
Speaker identification raises a separate question. Under the GDPR, biometric data processed to uniquely identify a person is a special category of data (Article 9(1)). Recordings that hold card details fall under the payment card rules described under call center analytics.
Frequently asked questions
What is the difference between speech analytics and voice analytics?
The terms are often used interchangeably. When they are separated, speech analytics means analyzing the words in a call. Voice analytics means analyzing the sound of the voice, such as pitch, pace, and volume.
Is speech analytics the same as call recording?
Call recording stores the audio. Speech analytics transcribes the recordings, or the live call, and finds patterns across many calls. Many speech analytics tools rely on a recording system as their source.
Can speech analytics detect customer emotions in the EU?
The Commission's guidelines say that tracking customers' emotions in a call center falls outside the AI Act's workplace ban. Tracking employees' emotions falls under it. The deployer has to inform customers that emotion recognition is in use (Article 50(3)). GDPR rules on personal data apply as well.
How accurate is speech analytics?
It depends mostly on transcription accuracy for your languages, accents, and phone lines. Published accuracy figures come from test sets that may not match your calls. A test on a sample of your own recordings, scored by word error rate and by how well topics are tagged, gives a usable number.
How we approach this at Zowie
Zowie builds AI agents for voice, chat, and email. On a call, the language model speaks. A separate, deterministic Decision Engine returns every amount, rate, offer, disclosure, and escalation from your own rules. Each decision is written to an audit log. You are welcome to book a demo.
Explore: Call center analytics, Sentiment analysis, Word error rate, AI quality monitoring, PII redaction, Voice AI