emotion_clustering.py

from sklearn.cluster import *
import os
import numpy as np
from config import config
import yaml
from multiprocessing import Pool
from tqdm import tqdm


def process_speaker(speaker):
    embs = []
    wavnames = []
    print("\nspeaker: " + speaker)
    for file in filelist_dict[speaker]:
        try:
            embs.append(
                np.expand_dims(np.load(f"{os.path.splitext(file)[0]}.emo.npy"), axis=0)
            )
            wavnames.append(os.path.basename(file))
        except Exception as e:
            print(e)
    x = np.concatenate(embs, axis=0)
    x = np.squeeze(x)
    # 聚类算法类的数量
    n_clusters = config.emo_cluster_config.n_clusters
    method = config.emo_cluster_config.method
    if method == "b":
        model = Birch(n_clusters=n_clusters, threshold=0.2)
    elif method == "s":
        model = SpectralClustering(n_clusters=n_clusters)
    elif method == "a":
        model = AgglomerativeClustering(n_clusters=n_clusters)
    else:
        model = KMeans(n_clusters=n_clusters, random_state=42)
    # 可以自行尝试各种不同的聚类算法
    y_predict = model.fit_predict(x)
    classes = [[] for i in range(y_predict.max() + 1)]

    for idx, wavname in enumerate(wavnames):
        classes[y_predict[idx]].append(wavname)

    yml_result = {}
    yml_result[speaker] = {}
    os.makedirs(
        os.path.join(config.dataset_path, f"emo_clustering/{speaker}"), exist_ok=True
    )
    for i in range(y_predict.max() + 1):
        class_length = len(classes[i])
        print("类别:", i, "本类中样本数量:", class_length)
        yml_result[speaker][f"class{i}"] = []
        for j in range(config.emo_cluster_config.n_samples):
            if j >= class_length:
                break
            print(classes[i][j])
            yml_result[speaker][f"class{i}"].append(classes[i][j])
        if hasattr(model, "cluster_centers_") and config.emo_cluster_config.save_center:
            centers = model.cluster_centers_
            filename = os.path.join(
                config.dataset_path, f"emo_clustering/{speaker}/cluster_center_{i}.npy"
            )
            # 保存中心
            np.save(filename, centers[i])
        elif config.emo_cluster_config.save_center:
            labels = model.labels_
            centers = np.array([X[labels == i].mean(0) for i in range(n_clusters)])
            filename = os.path.join(
                config.dataset_path, f"emo_clustering/{speaker}/cluster_center_{i}.npy"
            )
            np.save(filename, centers[i])
    return yml_result


if __name__ == "__main__":
    filelist_dict = {}
    with open(
        config.preprocess_text_config.train_path, mode="r", encoding="utf-8"
    ) as f:
        for line in f:
            speaker = line.split("|")[1]
            if speaker not in filelist_dict:
                filelist_dict[speaker] = []
            filelist_dict[speaker].append(line.split("|")[0])

    with Pool() as p:
        results = list(
            tqdm(
                p.imap(process_speaker, list(filelist_dict.keys())),
                total=len(filelist_dict),
            )
        )

    yml_result = {}
    for result in results:
        yml_result.update(result)

    with open(
        os.path.join(config.dataset_path, "emo_clustering/emo_clustering.yml"),
        "w",
        encoding="utf-8",
    ) as f:
        yaml.dump(yml_result, f)