AI glossary · Voice
Speech recognition
In one sentence
Speech recognition is the branch of artificial intelligence that lets a machine identify the words a person says and turn them into text or commands. Its technical name is ASR, short for automatic speech recognition.
Also known as: automatic speech recognition · ASR · voice recognition · speech-to-text recognition
What is speech recognition?
It’s the discipline of teaching machines to hear. It has been developing for decades: early systems recognised single digits from one speaker, and 1990s dictation software made you train it on your voice and speak slowly. Today’s models, based on neural networks trained on vast amounts of audio, understand almost anyone, with an accent and at normal speed.
Put it to work
A voice agent that picks up the phone
From €149/mo + VAT
It answers calls 24/7, books appointments and logs every call in the CRM.
See the voice agentIt helps to separate three ideas that sound alike:
| Term | Question it answers |
|---|---|
| Speech recognition (ASR) | Which words were said? |
| Voice (speaker) recognition | Who is speaking? |
| Natural language understanding (NLU) | What does the person want? |
When speech recognition is sold as a ready-to-use service, it’s called speech to text. When it’s applied to recordings to produce documents, it’s automatic transcription.
Related to this term
UTIO Voice
- 500 minutes and a number included
- Natural conversation in 26 languages
- Introduces itself as a virtual assistant
- 14-day free trial, no card needed
From €149/mo + VAT
See the voice agentHow it works
Modern speech recognition combines two kinds of knowledge:
- Acoustic. Which sounds match which pieces of words, learned from thousands of hours of speech paired with text.
- Linguistic. Which word sequences are likely in a language. That’s how it tells “their”, “there” and “they’re” apart from the rest of the sentence.
These days both usually sit inside one end-to-end model: audio goes in, text comes out. Quality is measured by word error rate (WER), which varies a lot with noise, microphone, accent and vocabulary. Phone audio is among the hardest, because the line strips out some of the voice’s frequencies.
Example
A farm supplies co-operative in rural Wales takes calls from members about orders and deliveries. Many have strong regional accents and use trade vocabulary. Testing a speech recognition system, the co-op finds it handles general sentences well but trips over product and breed names. The fix is to add those words as custom vocabulary and have the system repeat the key detail before moving on: “Just to confirm, that’s Texel?”
Speech recognition at UTIO
UTIO’s voice agent uses real-time speech recognition on every call. It understands everyday speech, interruptions and accents, and nobody has to press keys or slow down. The text it produces feeds the language model, which decides what to say and which action to take: book, qualify, take a message or transfer.
Transcription runs on UTIO’s own servers, and transcripts are stored in Spain alongside each call in the CRM. The agent converses in 26 languages.
Frequently asked questions
Is speech recognition the same as voice recognition?
People often use them interchangeably, but strictly speaking no. Speech recognition works out what is said. Voice recognition, or speaker recognition, works out who is speaking and is used to verify identity. They are different technologies.
Does recognising speech mean understanding the request?
No. Speech recognition turns sound into words. Understanding the intent behind them, for example that “I wanted to move the Tuesday thing” means rescheduling an appointment, is the job of natural language understanding and the language model.
Why does it get some names or numbers wrong?
Because the model picks the most likely sequence of words. An unusual surname or an order number doesn't follow common patterns. Results improve with custom word lists and by having the system read the detail back to confirm it.
Related terms
- Speech to text
Speech to text (STT) is the technology that turns spoken audio into written text. It's the software component, usually a model reached through an API, that does the listening in a voicebot, a dictation tool or a transcription service, either live or on recordings.
- Automatic transcription
Automatic transcription is the process of turning an audio or video recording into written text without anyone typing it, using speech recognition. It usually adds punctuation, timestamps and labels showing who is speaking at each point.
- Voicebot
A voicebot is a program that holds spoken conversations: it listens to what a person says, works out what they want and answers out loud. It can run on a phone line, a website, an app or a smart speaker, and with generative AI it handles open conversation, not just menus.
- Natural language understanding (NLU)
Natural language understanding (NLU) is the part of NLP that deals with the meaning of a message: what the person wants (their intent) and which specific details they give (entities such as dates, names or products).
- Neural network
An artificial neural network is a mathematical model made of layers of connected units, called neurons, that turn a numeric input into an output. It learns by adjusting the weight of each connection based on examples.
Updated on