提交

kongdeqiang

2026-03-13 28ccfbfc51068a663a80764e14074df5edf2b5ba

 funasr/models/campplus/cluster_backend.py

@@ -51,7 +51,7 @@

    def p_pruning(self, A):
        if A.shape[0] * self.pval < 6:
            pval = 6. / A.shape[0]
            pval = 6.0 / A.shape[0]
        else:
            pval = self.pval

@@ -80,7 +80,8 @@
            num_of_spk = k_oracle
        else:
            lambda_gap_list = self.getEigenGaps(
                lambdas[self.min_num_spks - 1:self.max_num_spks + 1])
                lambdas[self.min_num_spks - 1 : self.max_num_spks + 1]
            )
            num_of_spk = np.argmax(lambda_gap_list) + self.min_num_spks

        emb = eig_vecs[:, :num_of_spk]
@@ -105,12 +106,9 @@
      Emphasis On Topological Structure. ICASSP2022
    """

    def __init__(self,
                 n_neighbors=20,
                 n_components=60,
                 min_samples=10,
                 min_cluster_size=10,
                 metric='cosine'):
    def __init__(
        self, n_neighbors=20, n_components=60, min_samples=10, min_cluster_size=10, metric="cosine"
    ):
        self.n_neighbors = n_neighbors
        self.n_components = n_components
        self.min_samples = min_samples
@@ -119,6 +117,7 @@

    def __call__(self, X):
        import umap.umap_ as umap

        umap_X = umap.UMAP(
            n_neighbors=self.n_neighbors,
            min_dist=0.0,
@@ -128,7 +127,8 @@
        labels = HDBSCAN(
            min_samples=self.min_samples,
            min_cluster_size=self.min_cluster_size,
            allow_single_cluster=True).fit_predict(umap_X)
            allow_single_cluster=True,
        ).fit_predict(umap_X)
        return labels


@@ -139,9 +139,9 @@
        model_config: The model config.
    """

    def __init__(self):
    def __init__(self, merge_thr=0.78):
        super().__init__()
        self.model_config = {'merge_thr':0.78}
        self.model_config = {"merge_thr": merge_thr}
        # self.other_config = kwargs

        self.spectral_cluster = SpectralCluster()
@@ -149,21 +149,18 @@

    def forward(self, X, **params):
        # clustering and return the labels
        k = params['oracle_num'] if 'oracle_num' in params else None
        assert len(
            X.shape
        ) == 2, 'modelscope error: the shape of input should be [N, C]'
        k = params["oracle_num"] if "oracle_num" in params else None
        assert len(X.shape) == 2, "modelscope error: the shape of input should be [N, C]"
        if X.shape[0] < 20:
            return np.zeros(X.shape[0], dtype='int')
            return np.zeros(X.shape[0], dtype="int")
        if X.shape[0] < 2048 or k is not None:
            # unexpected corner case
            labels = self.spectral_cluster(X, k)
        else:
            labels = self.umap_hdbscan_cluster(X)

        if k is None and 'merge_thr' in self.model_config:
            labels = self.merge_by_cos(labels, X,
                                       self.model_config['merge_thr'])
        if k is None and "merge_thr" in self.model_config:
            labels = self.merge_by_cos(labels, X, self.model_config["merge_thr"])

        return labels

@@ -180,8 +177,7 @@
                spk_center.append(spk_emb)
            assert len(spk_center) > 0
            spk_center = np.stack(spk_center, axis=0)
            norm_spk_center = spk_center / np.linalg.norm(
                spk_center, axis=1, keepdims=True)
            norm_spk_center = spk_center / np.linalg.norm(spk_center, axis=1, keepdims=True)
            affinity = np.matmul(norm_spk_center, norm_spk_center.T)
            affinity = np.triu(affinity, 1)
            spks = np.unravel_index(np.argmax(affinity), affinity.shape)