dtw-speech-aligner

A Python tool for speech segment alignment using Dynamic Time Warping (DTW). Assumes input audio has been pre‑processed by VAD (voice activity detection).

Features

• Subsequence DTW alignment of two speech segments (query vs. reference)
• Supports MFCC and/or fundamental‑frequency (F0) features
• Outputs clipped reference audio and optional diagnostic plots

Usage

python main.py \
  --query_path   path/to/query_audio \
  --reference_path path/to/reference_audio \
  [--feat_types mfcc f0] \
  [--save_plot]

• query_path: path to the (pre‑VAD) query audio.
• reference_path: path to the (pre‑VAD) reference audio.
• feat_types: which features to use: mfcc, f0 (default: mfcc).
• save_plot: save DTW & spectrogram plots.

Examples

Chinese

Audio

Query (TTS): query_chinese.wav

Reference (Human): reference_chinese.wav

Clipped segment (DTW): clip_chinese.wav

Visualization

Alignment

Mel-spectrogram & f0

MFCC DTW paths

English

Audio

Query (TTS): query_english.wav

Reference (Human): reference_english.wav

Clipped segment (DTW): clip_english.wav

Visualization

Alignment

Mel-spectrogram & f0

MFCC DTW paths

Taiwanese

Audio

Query (TTS): query_taiwanese.wav

Reference (Human): reference_taiwanese.mp3

Clipped segment (DTW): clip_taiwanese.wav

Visualization

Alignment

Mel-spectrogram & f0

MFCC DTW paths