code/extract_xlsr.py

#!/usr/bin/env python3
"""Extract XLS-R (wav2vec2 multilingual) embeddings.

Model: facebook/wav2vec2-xls-r-300m (self-supervised multilingual, 1024-dim)
Frame-level output is mean-pooled to get utterance-level embeddings.
Install: pip install transformers
"""

import argparse
import torch
import numpy as np
from extraction_utils import load_audio, extract_all


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--device", default="cuda" if torch.cuda.is_available() else "cpu")
    parser.add_argument("--base-dir", default=None)
    parser.add_argument("--output-dir", default=None)
    args = parser.parse_args()

    from transformers import Wav2Vec2Model, Wav2Vec2FeatureExtractor

    print(f"Loading XLS-R 300M on {args.device}...")
    feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("facebook/wav2vec2-xls-r-300m")
    model = Wav2Vec2Model.from_pretrained("facebook/wav2vec2-xls-r-300m").to(args.device)
    model.eval()

    def model_fn(audio_path):
        audio = load_audio(audio_path, target_sr=16000)
        inputs = feature_extractor(audio, sampling_rate=16000, return_tensors="pt")
        inputs = {k: v.to(args.device) for k, v in inputs.items()}
        with torch.no_grad():
            outputs = model(**inputs)
        hidden = outputs.last_hidden_state  # (1, T, 1024)
        embedding = hidden.mean(dim=1).squeeze().cpu().numpy()
        return embedding

    extract_all(model_fn, "xlsr", args.base_dir, args.output_dir)


if __name__ == "__main__":
    main()