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
auto frontend
游雁e9acc5db07da
diff --git a/examples/industrial_data_pretraining/llm_asr/conf/whisper_qwen_linear2.yaml b/examples/industrial_data_pretraining/llm_asr/conf/whisper_qwen_linear2.yamlindex 59e93a6c9..62d77e260 100644--- a/examples/industrial_data_pretraining/llm_asr/conf/whisper_qwen_linear2.yaml+++ b/examples/industrial_data_pretraining/llm_asr/conf/whisper_qwen_linear2.yaml@@ -35,7 +35,7 @@ frontend: WhisperFrontendfrontend_conf:fs: 16000whisper_model: large-v3- do_pad_trim: true+ do_pad_trim: falsepermute: false # true: [bs, frames, dims]; false: [bs, dims, frames]filters_path: "/nfs/zhifu.gzf/init_model/SenseVoiceModelscope/assets/mel_filters.npz"diff --git a/funasr/datasets/openai_datasets/datasets.py b/funasr/datasets/openai_datasets/datasets.pyindex 9a542adb6..9bd0698fb 100644--- a/funasr/datasets/openai_datasets/datasets.py+++ b/funasr/datasets/openai_datasets/datasets.py@@ -123,21 +123,20 @@ class OpenAIDataset(torch.utils.data.Dataset):) # speech: [b, T, d]if self.permute:speech = speech.permute(0, 2, 1)- if speech_lengths > self.batch_size:- continue+ # if speech_lengths > self.batch_size:+ # continue- fbank_lens = speech_lengths[0].item()- olens = 1 + (fbanks_len - 3 + 2 * 1) // 2+ olens = 1 + (speech_lengths[0].item() - 3 + 2 * 1) // 2olens = 1 + (olens - 3 + 2 * 1) // 2sub_token_len = (olens - 1) // 2 + 1- sub_token = [0] * sub_token_len[0]+ sub_token = [0] * sub_token_lenfbank_beg_i = [len(source_ids)]source_ids += sub_tokenfbank_mask_i += [1] * len(sub_token)source_mask = [-100] * len(source_ids)target_out = f"{target_out}<|im_end|>"- target_ids = tokenizer.encode(target_out)+ target_ids = self.tokenizer.encode(target_out)input_ids += source_ids + target_idslabels += source_mask + target_idsfbank_mask += fbank_mask_idiff --git a/funasr/models/llm_asr/model.py b/funasr/models/llm_asr/model.pyindex 11db0096d..411b59d94 100644--- a/funasr/models/llm_asr/model.py+++ b/funasr/models/llm_asr/model.py@@ -385,13 +385,6 @@ class LLMASR2(nn.Module):super().__init__()- if specaug is not None:- specaug_class = tables.specaug_classes.get(specaug)- specaug = specaug_class(**specaug_conf)- if normalize is not None:- normalize_class = tables.normalize_classes.get(normalize)- normalize = normalize_class(**normalize_conf)-# audio encoderhub = audio_encoder_conf.get("hub", None)if hub == "ms":@@ -422,23 +415,23 @@ class LLMASR2(nn.Module):# llmhub = llm_conf.get("hub", "hf")self.llm = None- # if hub == "hf":- # from transformers import AutoModelForCausalLM, AutoTokenizer, AutoConfig- #- # init_param_path = llm_conf.get("init_param_path", "vicuna-7b-v1.5")- #- # model = AutoModelForCausalLM.from_pretrained(- # init_param_path,- # load_in_8bit=None,- # device_map=None,- # use_cache=None,- # )- # freeze = llm_conf.get("freeze", True)- # if freeze:- # for name, param in model.named_parameters():- # param.requires_grad = False- # model.eval()- # self.llm = model+ if hub == "hf":+ from transformers import AutoModelForCausalLM, AutoTokenizer, AutoConfig++ init_param_path = llm_conf.get("init_param_path", "vicuna-7b-v1.5")++ model = AutoModelForCausalLM.from_pretrained(+ init_param_path,+ load_in_8bit=None,+ device_map=None,+ use_cache=None,+ )+ freeze = llm_conf.get("freeze", True)+ if freeze:+ for name, param in model.named_parameters():+ param.requires_grad = False+ model.eval()+ self.llm = model# adaptoradaptor_class = tables.adaptor_classes.get(audio_adaptor)@@ -447,21 +440,6 @@ class LLMASR2(nn.Module):self.audio_adaptor = audio_adaptor- self.blank_id = blank_id- self.sos = sos if sos is not None else vocab_size - 1- self.eos = eos if eos is not None else vocab_size - 1- self.vocab_size = vocab_size- self.ignore_id = ignore_id- self.specaug = specaug- self.normalize = normalize-- self.criterion_att = LabelSmoothingLoss(- size=vocab_size,- padding_idx=ignore_id,- smoothing=lsm_weight,- normalize_length=length_normalized_loss,- )-self.error_calculator = Noneself.length_normalized_loss = length_normalized_loss@@ -493,10 +471,10 @@ class LLMASR2(nn.Module):batch_size = speech.shape[0]# audio encoder- encoder_out, encoder_out_lens = self.encode(speech, speech_lengths)+ encoder_out, encoder_out_lens = self.audio_encoder(speech.permute(0, 2, 1), speech_lengths)# audio_adaptor- encoder_out = self.audio_adaptor(encoder_out)+ encoder_out, encoder_out_lens = self.audio_adaptor(encoder_out, encoder_out_lens)input_ids[input_ids == -1] = 0input_ids[input_ids == -100] = 0@@ -530,24 +508,10 @@ class LLMASR2(nn.Module):# force_gatherable: to-device and to-tensor if scalar for DataParallelif self.length_normalized_loss:- batch_size = int((text_lengths + 1).sum())+ batch_size = int((labels_ids > 0 + 1).sum())loss, stats, weight = force_gatherable((loss, stats, batch_size), loss.device)return loss, stats, weight- def encode(- self,- speech: torch.Tensor,- speech_lengths: torch.Tensor,- **kwargs,- ):- speech = speech.permute(0, 2, 1)- res = self.audio_encoder(speech)- if isinstance(res, (list, tuple)):- encoder_out, encoder_out_lens = res[0], res[1]- else:- encoder_out, encoder_out_lens = res, speech_lengths- return encoder_out, encoder_out_lens-def inference(self,data_in,