Applications · Established · Beginner

Speech Recognition

Also known as: ASR, Speech-to-text

Converting spoken audio into written text, optionally with speaker labels and timestamps.

What Speech Recognition is

Modern ASR is robust to accents, background noise and code-switching in a way earlier systems were not, and often runs multilingual out of the box.

How it works

Audio is turned into spectrogram features and processed by an encoder-decoder or transducer model trained on very large speech corpora. Diarisation separates speakers, and punctuation and casing are restored by a post-processing model.

Why it matters

It underpins meeting notes, captioning, voice interfaces and the vast market for call analytics.

Common uses

→Meeting transcription and summaries
→Captioning and subtitles
→Voice commands
→Call-centre quality analysis

Strengths

✓Near-human accuracy on clear speech
✓Multilingual
✓Runs locally in small variants

Watch for

✓Degrades on heavy noise and overlapping speakers
✓Domain terminology errors
✓Recording consent obligations

Continue exploring

Speech Synthesis Conversational AI Multimodal AI