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
bugfix memory leaky
游雁fc547e14e818
diff --git a/funasr/models/lcbnet/attention.py b/funasr/models/lcbnet/attention.pyindex 05a5041f2..83753ede0 100644--- a/funasr/models/lcbnet/attention.py+++ b/funasr/models/lcbnet/attention.py@@ -78,19 +78,19 @@ class MultiHeadedAttentionReturnWeight(nn.Module):mask = mask.unsqueeze(1).eq(0) # (batch, 1, *, time2)min_value = torch.finfo(scores.dtype).minscores = scores.masked_fill(mask, min_value)- self.attn = torch.softmax(scores, dim=-1).masked_fill(+ attn = torch.softmax(scores, dim=-1).masked_fill(mask, 0.0) # (batch, head, time1, time2)else:- self.attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)+ attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)- p_attn = self.dropout(self.attn)+ p_attn = self.dropout(attn)x = torch.matmul(p_attn, value) # (batch, head, time1, d_k)x = (x.transpose(1, 2).contiguous().view(n_batch, -1, self.h * self.d_k)) # (batch, time1, d_model)- return self.linear_out(x), self.attn # (batch, time1, d_model)+ return self.linear_out(x), attn # (batch, time1, d_model)def forward(self, query, key, value, mask):"""Compute scaled dot product attention.diff --git a/funasr/models/sanm/multihead_att.py b/funasr/models/sanm/multihead_att.pyindex c7d97961d..671d46011 100644--- a/funasr/models/sanm/multihead_att.py+++ b/funasr/models/sanm/multihead_att.py@@ -55,8 +55,8 @@ class MultiHeadedAttentionSANMExport(nn.Module):def forward_attention(self, value, scores, mask):scores = scores + mask- self.attn = torch.softmax(scores, dim=-1)- context_layer = torch.matmul(self.attn, value) # (batch, head, time1, d_k)+ attn = torch.softmax(scores, dim=-1)+ context_layer = torch.matmul(attn, value) # (batch, head, time1, d_k)context_layer = context_layer.permute(0, 2, 1, 3).contiguous()new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)@@ -134,8 +134,8 @@ class MultiHeadedAttentionCrossAttExport(nn.Module):def forward_attention(self, value, scores, mask):scores = scores + mask- self.attn = torch.softmax(scores, dim=-1)- context_layer = torch.matmul(self.attn, value) # (batch, head, time1, d_k)+ attn = torch.softmax(scores, dim=-1)+ context_layer = torch.matmul(attn, value) # (batch, head, time1, d_k)context_layer = context_layer.permute(0, 2, 1, 3).contiguous()new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)@@ -177,8 +177,8 @@ class OnnxMultiHeadedAttention(nn.Module):def forward_attention(self, value, scores, mask):scores = scores + mask- self.attn = torch.softmax(scores, dim=-1)- context_layer = torch.matmul(self.attn, value) # (batch, head, time1, d_k)+ attn = torch.softmax(scores, dim=-1)+ context_layer = torch.matmul(attn, value) # (batch, head, time1, d_k)context_layer = context_layer.permute(0, 2, 1, 3).contiguous()new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)@@ -232,8 +232,8 @@ class OnnxRelPosMultiHeadedAttention(OnnxMultiHeadedAttention):def forward_attention(self, value, scores, mask):scores = scores + mask- self.attn = torch.softmax(scores, dim=-1)- context_layer = torch.matmul(self.attn, value) # (batch, head, time1, d_k)+ attn = torch.softmax(scores, dim=-1)+ context_layer = torch.matmul(attn, value) # (batch, head, time1, d_k)context_layer = context_layer.permute(0, 2, 1, 3).contiguous()new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)diff --git a/funasr/models/sense_voice/model.py b/funasr/models/sense_voice/model.pyindex 1311987f2..25e9fafd3 100644--- a/funasr/models/sense_voice/model.py+++ b/funasr/models/sense_voice/model.py@@ -196,13 +196,13 @@ class MultiHeadedAttentionSANM(nn.Module):"inf") # float(numpy.finfo(torch.tensor(0, dtype=scores.dtype).numpy().dtype).min)scores = scores.masked_fill(mask, min_value)- self.attn = torch.softmax(scores, dim=-1).masked_fill(+ attn = torch.softmax(scores, dim=-1).masked_fill(mask, 0.0) # (batch, head, time1, time2)else:- self.attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)+ attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)- p_attn = self.dropout(self.attn)+ p_attn = self.dropout(attn)x = torch.matmul(p_attn, value) # (batch, head, time1, d_k)x = (x.transpose(1, 2).contiguous().view(n_batch, -1, self.h * self.d_k)@@ -644,7 +644,13 @@ class SenseVoiceSmall(nn.Module):self.embed = torch.nn.Embedding(7 + len(self.lid_dict) + len(self.textnorm_dict), input_size)- self.emo_dict = {"unk": 25009, "happy": 25001, "sad": 25002, "angry": 25003, "neutral": 25004}+ self.emo_dict = {+ "unk": 25009,+ "happy": 25001,+ "sad": 25002,+ "angry": 25003,+ "neutral": 25004,+ }self.criterion_att = LabelSmoothingLoss(size=self.vocab_size,@@ -874,7 +880,7 @@ class SenseVoiceSmall(nn.Module):ctc_logits = self.ctc.log_softmax(encoder_out)if kwargs.get("ban_emo_unk", False):ctc_logits[:, :, self.emo_dict["unk"]] = -float("inf")-+results = []b, n, d = encoder_out.size()if isinstance(key[0], (list, tuple)):diff --git a/funasr/models/sond/attention.py b/funasr/models/sond/attention.pyindex 1af553477..18580b7fe 100644--- a/funasr/models/sond/attention.py+++ b/funasr/models/sond/attention.py@@ -84,13 +84,13 @@ class MultiHeadedAttention(nn.Module):mask = mask.unsqueeze(1).eq(0) # (batch, 1, *, time2)min_value = float(numpy.finfo(torch.tensor(0, dtype=scores.dtype).numpy().dtype).min)scores = scores.masked_fill(mask, min_value)- self.attn = torch.softmax(scores, dim=-1).masked_fill(+ attn = torch.softmax(scores, dim=-1).masked_fill(mask, 0.0) # (batch, head, time1, time2)else:- self.attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)+ attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)- p_attn = self.dropout(self.attn)+ p_attn = self.dropout(attn)x = torch.matmul(p_attn, value) # (batch, head, time1, d_k)x = (x.transpose(1, 2).contiguous().view(n_batch, -1, self.h * self.d_k)@@ -287,13 +287,13 @@ class MultiHeadSelfAttention(nn.Module):min_value = float(numpy.finfo(torch.tensor(0, dtype=scores.dtype).numpy().dtype).min)scores = scores.masked_fill(mask, min_value)- self.attn = torch.softmax(scores, dim=-1).masked_fill(+ attn = torch.softmax(scores, dim=-1).masked_fill(mask, 0.0) # (batch, head, time1, time2)else:- self.attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)+ attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)- p_attn = self.dropout(self.attn)+ p_attn = self.dropout(attn)x = torch.matmul(p_attn, value) # (batch, head, time1, d_k)x = (x.transpose(1, 2).contiguous().view(n_batch, -1, self.h * self.d_k)diff --git a/funasr/models/transformer/attention.py b/funasr/models/transformer/attention.pyindex 6e6f75411..2844333a1 100644--- a/funasr/models/transformer/attention.py+++ b/funasr/models/transformer/attention.py@@ -87,13 +87,13 @@ class MultiHeadedAttention(nn.Module):"inf") # min_value = float(np.finfo(torch.tensor(0, dtype=qk.dtype).numpy().dtype).min)scores = scores.masked_fill(mask, min_value)- self.attn = torch.softmax(scores, dim=-1).masked_fill(+ attn = torch.softmax(scores, dim=-1).masked_fill(mask, 0.0) # (batch, head, time1, time2)else:- self.attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)+ attn = torch.softmax(scores, dim=-1) # (batch, head, time1, time2)- p_attn = self.dropout(self.attn)+ p_attn = self.dropout(attn)x = torch.matmul(p_attn, value) # (batch, head, time1, d_k)x = (x.transpose(1, 2).contiguous().view(n_batch, -1, self.h * self.d_k)@@ -154,8 +154,8 @@ class MultiHeadedAttentionExport(nn.Module):def forward_attention(self, value, scores, mask):scores = scores + mask- self.attn = torch.softmax(scores, dim=-1)- context_layer = torch.matmul(self.attn, value) # (batch, head, time1, d_k)+ attn = torch.softmax(scores, dim=-1)+ context_layer = torch.matmul(attn, value) # (batch, head, time1, d_k)context_layer = context_layer.permute(0, 2, 1, 3).contiguous()new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)@@ -209,8 +209,8 @@ class RelPosMultiHeadedAttentionExport(MultiHeadedAttentionExport):def forward_attention(self, value, scores, mask):scores = scores + mask- self.attn = torch.softmax(scores, dim=-1)- context_layer = torch.matmul(self.attn, value) # (batch, head, time1, d_k)+ attn = torch.softmax(scores, dim=-1)+ context_layer = torch.matmul(attn, value) # (batch, head, time1, d_k)context_layer = context_layer.permute(0, 2, 1, 3).contiguous()new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)@@ -575,9 +575,9 @@ class RelPositionMultiHeadedAttentionChunk(torch.nn.Module):if chunk_mask is not None:mask = chunk_mask.unsqueeze(0).unsqueeze(1) | maskscores = scores.masked_fill(mask, float("-inf"))- self.attn = torch.softmax(scores, dim=-1).masked_fill(mask, 0.0)+ attn = torch.softmax(scores, dim=-1).masked_fill(mask, 0.0)- attn_output = self.dropout(self.attn)+ attn_output = self.dropout(attn)attn_output = torch.matmul(attn_output, value)attn_output = self.linear_out(