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
bug fix
维石487189b949a2
diff --git a/runtime/python/libtorch/demo_contextual_paraformer.py b/runtime/python/libtorch/demo_contextual_paraformer.pyindex 06c0f7619..306981c0a 100644--- a/runtime/python/libtorch/demo_contextual_paraformer.py+++ b/runtime/python/libtorch/demo_contextual_paraformer.py@@ -7,7 +7,7 @@ device_id = 0 if torch.cuda.is_available() else -1model = ContextualParaformer(model_dir, batch_size=1, device_id=device_id) # gpuwav_path = "{}/.cache/modelscope/hub/{}/example/asr_example.wav".format(Path.home(), model_dir)-hotwords = "你的热词 魔哒"+hotwords = "你的热词 魔搭"result = model(wav_path, hotwords)print(result)diff --git a/runtime/python/libtorch/demo_seaco_paraformer.py b/runtime/python/libtorch/demo_seaco_paraformer.pyindex d54dce537..ad28bfe97 100644--- a/runtime/python/libtorch/demo_seaco_paraformer.py+++ b/runtime/python/libtorch/demo_seaco_paraformer.py@@ -7,7 +7,7 @@ device_id = 0 if torch.cuda.is_available() else -1model = SeacoParaformer(model_dir, batch_size=1, device_id=device_id) # gpuwav_path = "{}/.cache/modelscope/hub/{}/example/asr_example.wav".format(Path.home(), model_dir)-hotwords = "你的热词 魔哒"+hotwords = "你的热词 魔搭"result = model(wav_path, hotwords)print(result)diff --git a/runtime/python/libtorch/funasr_torch/paraformer_bin.py b/runtime/python/libtorch/funasr_torch/paraformer_bin.pyindex 9f35db718..755237e40 100644--- a/runtime/python/libtorch/funasr_torch/paraformer_bin.py+++ b/runtime/python/libtorch/funasr_torch/paraformer_bin.py@@ -316,7 +316,10 @@ class ContextualParaformer(Paraformer):) -> List:# make hotword listhotwords, hotwords_length = self.proc_hotword(hotwords)- bias_embed = self.eb_infer(torch.Tensor(hotwords))+ if int(self.device_id) != -1:+ bias_embed = self.eb_infer(hotwords.cuda())+ else:+ bias_embed = self.eb_infer(hotwords)# index from bias_embedbias_embed = torch.transpose(bias_embed, 0, 1)_ind = np.arange(0, len(hotwords)).tolist()@@ -334,7 +337,7 @@ class ContextualParaformer(Paraformer):outputs = self.bb_infer(feats, feats_len, bias_embed)am_scores, valid_token_lens = outputs[0], outputs[1]else:- outputs = self.bb_infer_infer(feats.cuda(), feats_len.cuda(), bias_embed)+ outputs = self.bb_infer_infer(feats.cuda(), feats_len.cuda(), bias_embed.cuda())am_scores, valid_token_lens = outputs[0].cpu(), outputs[1].cpu()except:# logging.warning(traceback.format_exc())@@ -369,7 +372,7 @@ class ContextualParaformer(Paraformer):hotword_int = [word_map(i) for i in hotwords]hotword_int.append(np.array([1]))hotwords = pad_list(hotword_int, pad_value=0, max_len=10)- return hotwords, hotwords_length+ return torch.tensor(hotwords), hotwords_lengthdef bb_infer(self, feats, feats_len, bias_embed