Skip to content
Glossary

Speech recognition (ASR)

Speech recognition, also called automatic speech recognition or ASR, is technology that turns spoken words into written text. On a phone call it converts the caller's voice into a transcript as they speak, so software such as an AI voice agent can understand the request and respond.

How speech recognition works on Indian calls

A speech recognition model listens to short slices of audio and predicts the words in them, revising its guess as more audio arrives. On a live call it streams, so the transcript builds while the caller is still talking.

Indian phone calls make the job harder in a few specific ways:

  • Phone-line audio. A call carries less of the voice than a good microphone, so similar sounds blur together.
  • Background noise. Traffic, a shop counter or a TV in the next room.
  • Mixed languages. "Mujhe ek personal loan chahiye, around five lakh" changes language halfway through.
  • Numbers said many ways. "Dhai hazaar", "pachchees sau" and "two thousand five hundred" all mean ₹2,500.
  • Names and places. Surnames, colony names and brand names are easy to mishear.

Why it matters

Everything an agent does rests on hearing the caller correctly. A misheard amount becomes a wrong promise to pay; a misheard date becomes a missed appointment. Good agents read key details back to the caller to confirm them, so slips get caught on the call instead of reaching your CRM.

How Kallix uses it

Speech recognition runs on every Kallix call and is included in the ₹4.80 a minute rate. Each call's transcript is saved with its recording, so you can read exactly what the agent heard. Before launch, test the agent with the phrases your customers really use, not textbook sentences.

See transcripts in call logs

Common questions.

More on the FAQ page, or write to us.

In practice, no. ASR is the technical name for the field; speech-to-text (STT) usually names the product or API that does the work.

It can, when the model has learned from Indian speech and mixed-language calls. Kallix agents work in Hindi, English, Hinglish and more, and follow the caller when they switch.

Usually by word error rate (WER): the share of words that are wrong, missing or extra compared with a careful human transcript. For calls, also check the words that matter most, such as names, dates and amounts.

Put an agent on the line.
Hear it call you today.

Hear a live callBook a demo

₹500 in free credits when you sign up. ₹4.80 a minute after that.