What Is Automatic Speech Recognition (ASR)?

The short version

Automatic Speech Recognition, or ASR, is the technology that listens to audio and writes down what was said. Your phone's dictation, meeting transcripts, and subtitles on videos all rely on it.

Whisper.cpp is a fast, local implementation of OpenAI's Whisper model. It does the same job as cloud speech APIs, but it runs entirely on your own machine — no audio ever leaves your computer.

How it works, in one picture

Audio preprocessing turns raw sound waves into a spectrogram (a picture of frequencies over time).
Acoustic model maps chunks of that spectrogram to sounds and word pieces.
Language model picks the most likely sentence out of many possible word sequences.
The result: readable text, often with punctuation and timestamps.

Why this matters for whisper.cpp

Whisper.cpp bundles all three steps into a single command-line tool written in C/C++. No Python runtime, no GPU required (though it can use one), and no internet connection.

# Transcribe a local audio file
./whisper-cli -m models/ggml-base.en.bin -f audio.wav

That single line does everything shown in the diagram above: it loads the model, preprocesses audio.wav, and prints the transcript to your terminal.

Common use cases

Turning podcast episodes into blog posts
Generating subtitles for videos
Creating searchable meeting notes
Building voice commands for local apps

Error handling basics

Even a "beginner" command needs a couple of safety checks:

if [ ! -f "$AUDIO_FILE" ]; then
  echo "Error: audio file not found: $AUDIO_FILE" >&2
  exit 1
fi
 
if [ ! -f "$MODEL_FILE" ]; then
  echo "Error: model file not found: $MODEL_FILE. Download one with models/download-ggml-model.sh" >&2
  exit 1
fi
 
./whisper-cli -m "$MODEL_FILE" -f "$AUDIO_FILE" || {
  echo "Error: whisper-cli exited with a non-zero status" >&2
  exit 1
}

Up next

The next article in this series covers how to actually build a small ASR pipeline with whisper.cpp — picking a model, converting audio formats, and reading the output.

The short version#

How it works, in one picture#

Why this matters for whisper.cpp#

Common use cases#

Error handling basics#