What is Speech-to-Text (ASR)?
Speech-to-text, also called automatic speech recognition (ASR), converts spoken audio into written text. In a voice agent it is the listening step that every later decision depends on.
Also known as: automatic speech recognition, ASR, STT, speech recognition
How it works
An ASR model maps audio frames to likely words. For live calls it runs in streaming mode, emitting partial transcripts while the person is still talking so the rest of the system can start work early. Telephone audio is typically narrowband, far thinner than studio audio, so models trained only on clean recordings often degrade on real calls.
Why it matters for voice agents
Everything downstream inherits recognition mistakes. If 'fifteen' becomes 'fifty' in an EMI amount, no language model downstream can know. Accuracy is therefore measured as word error rate, and it varies with accent, language, background noise and domain vocabulary such as drug names or loan terms. A single headline accuracy figure says little about your callers.
Common mistake
Evaluating ASR on a vendor's demo clip. Test on your own recorded calls, including noisy ones and mixed-language ones, and compare error rates on the numbers, names and dates your workflow actually depends on.
Where this shows up at Vaami
Vaami reports an aggregate word error rate under 6% across its production call corpus; per-language figures are not published.
See it on a real call.
Hear how a Vaami voice agent handles this on a live conversation in your language and your use case.