Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
Emo2Vec限定选择的情感类别 (#1730)
gaochangfeng2f7dcbad90e8
diff --git a/funasr/models/emotion2vec/model.py b/funasr/models/emotion2vec/model.pyindex 48b87160f..d18e1844c 100644--- a/funasr/models/emotion2vec/model.py+++ b/funasr/models/emotion2vec/model.py@@ -249,10 +249,17 @@ class Emotion2vec(torch.nn.Module):if self.proj:x = x.mean(dim=1)x = self.proj(x)+ for idx, lab in enumerate(labels):+ x[:,idx] = -np.inf if lab.startswith("unuse") else x[:,idx]x = torch.softmax(x, dim=-1)scores = x[0].tolist()- result_i = {"key": key[i], "labels": labels, "scores": scores}+ select_label = [lb for lb in labels if not lb.startswith("unuse")]+ select_score = [scores[idx] for idx, lb in enumerate(labels) if not lb.startswith("unuse")]++ # result_i = {"key": key[i], "labels": labels, "scores": scores}+ result_i = {"key": key[i], "labels": select_label, "scores": select_score}+if extract_embedding:result_i["feats"] = featsresults.append(result_i)