Day 1
Audio feature explorer
From sound to the features every speech model eats: waveform, spectrogram, log-mel, pitch.
Day 2
Your first speech model
A CNN that classifies Khmer sentences and speakers from log-mel.
Day 3
Speech recognition with CTC
A tiny CTC model from scratch, then Whisper and a Khmer ASR judge, scored by CER.
Day 4
Who is speaking?
Speaker embeddings, cosine trials, EER and DET curves: the attacker's view.
Day 5
Self-supervised speech encoders
wav2vec 2.0, HuBERT, WavLM: what each layer knows about who vs what.
Day 6
Making speech
The phase problem, vocoders, kNN-VC by hand, and the McAdams anonymizer.
Day 7
The full picture
A voice-privacy benchmark: privacy vs utility for every system.