Back to Projects

Build from first principles

Detect Speech in Noisy Audio

Turn sound into frames and a spectrogram, find where people speak, and keep working when background noise arrives.

What is voice activity detection?See how a program can tell when someone is talking

1Frame Energy in DecibelsHow loud is each moment?

2Loudness Threshold with a HangoverA first speech detector

3Precision, Recall and F1 per FrameScore a detector

4Build a Mel FilterbankHear frequencies like an ear

5Log-Mel SpectrogramFeatures a network can learn from

6From Frame Decisions to Speech SegmentsStart and end times

From Frames to a Speech DetectorSee how your functions fit together before the notebook

Build a Speech Detector That Survives NoiseBreak a loudness detector with background noise, then train one that keeps working