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
Dev gzf (#1377)
zhifu gaocdca62d933c4
diff --git a/examples/industrial_data_pretraining/paraformer/infer_demo.sh b/examples/industrial_data_pretraining/paraformer/demo.shsimilarity index 100%rename from examples/industrial_data_pretraining/paraformer/infer_demo.shrename to examples/industrial_data_pretraining/paraformer/demo.shdiff --git a/examples/industrial_data_pretraining/paraformer/finetune.sh b/examples/industrial_data_pretraining/paraformer/finetune.shindex 8bdd8daaf..266346cca 100644--- a/examples/industrial_data_pretraining/paraformer/finetune.sh+++ b/examples/industrial_data_pretraining/paraformer/finetune.sh@@ -6,7 +6,7 @@#git clone https://www.modelscope.cn/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch.git ${local_path}## generate jsonl from wav.scp and text.txt-#python funasr/datasets/audio_datasets/scp2jsonl.py \+#python -m funasr.datasets.audio_datasets.scp2jsonl \#++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \#++data_type_list='["source", "target"]' \#++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonldiff --git a/funasr/datasets/audio_datasets/scp2jsonl.py b/funasr/datasets/audio_datasets/scp2jsonl.pyindex b6df34ae3..e09a84a61 100644--- a/funasr/datasets/audio_datasets/scp2jsonl.py+++ b/funasr/datasets/audio_datasets/scp2jsonl.py@@ -72,14 +72,7 @@ def parse_context_length(data_list: list, data_type: str):@hydra.main(config_name=None, version_base=None)def main_hydra(cfg: DictConfig):- """- python funasr/datasets/audio_datasets/scp2jsonl.py \- ++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \- ++data_type_list='["source", "target"]' \- ++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl-- """-+kwargs = OmegaConf.to_container(cfg, resolve=True)scp_file_list = kwargs.get("scp_file_list", ("/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"))@@ -90,6 +83,13 @@ def main_hydra(cfg: DictConfig):gen_jsonl_from_wav_text_list(scp_file_list, data_type_list=data_type_list, jsonl_file_out=jsonl_file_out)+"""+python -m funasr.datasets.audio_datasets.scp2jsonl \+++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \+++data_type_list='["source", "target"]' \+++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl+"""+if __name__ == "__main__":main_hydra()diff --git a/funasr/models/mossformer/mossformer_encoder.py b/funasr/models/mossformer/mossformer_encoder.pyindex d06af999d..a28c960e8 100644--- a/funasr/models/mossformer/mossformer_encoder.py+++ b/funasr/models/mossformer/mossformer_encoder.py@@ -4,7 +4,7 @@ import torch.nn.functional as Ftry:from rotary_embedding_torch import RotaryEmbeddingexcept:- print("Please install rotary_embedding_torch by: \n pip install -U rotary_embedding_torch")+ print("If you want use mossformer, lease install rotary_embedding_torch by: \n pip install -U rotary_embedding_torch")from funasr.models.transformer.layer_norm import GlobalLayerNorm, CumulativeLayerNorm, ScaleNormfrom funasr.models.transformer.embedding import ScaledSinuEmbeddingfrom funasr.models.transformer.mossformer import FLASH_ShareA_FFConvMdiff --git a/funasr/models/paraformer/model.py b/funasr/models/paraformer/model.pyindex 729b8f500..90ce162d4 100644--- a/funasr/models/paraformer/model.py+++ b/funasr/models/paraformer/model.py@@ -455,7 +455,9 @@ class Paraformer(torch.nn.Module):speech, speech_lengths = data_in, data_lengthsif len(speech.shape) < 3:speech = speech[None, :, :]- if speech_lengths is None:+ if speech_lengths is not None:+ speech_lengths = speech_lengths.squeeze(-1)+ else:speech_lengths = speech.shape[1]else:# extract fbank featsdiff --git a/funasr/train_utils/trainer.py b/funasr/train_utils/trainer.pyindex c2326424f..6a59f91a1 100644--- a/funasr/train_utils/trainer.py+++ b/funasr/train_utils/trainer.py@@ -181,7 +181,7 @@ class Trainer:time2 = time.perf_counter()time_escaped = (time2 - time1)/3600.0- print(f"\ntime_escaped_epoch: {time_escaped:.3f} hours, estimated to finish {self.max_epoch} epoch: {(self.max_epoch-epoch)*time_escaped:.3f}\n")+ print(f"\nrank: {self.local_rank}, time_escaped_epoch: {time_escaped:.3f} hours, estimated to finish {self.max_epoch} epoch: {(self.max_epoch-epoch)*time_escaped:.3f}\n")if self.rank == 0:average_checkpoints(self.output_dir, self.avg_nbest_model)@@ -302,17 +302,14 @@ class Trainer:)pbar.set_description(description)if self.writer:- self.writer.add_scalar(f'rank{self.local_rank}_Loss/train', loss.item(),- epoch*len(self.dataloader_train) + batch_idx)+ self.writer.add_scalar(f'rank{self.local_rank}_Loss/train', loss.item(), self.batch_total)+ self.writer.add_scalar(f'rank{self.local_rank}_lr/train', lr, self.batch_total)for key, var in stats.items():- self.writer.add_scalar(f'rank{self.local_rank}_{key}/train', var.item(),- epoch * len(self.dataloader_train) + batch_idx)+ self.writer.add_scalar(f'rank{self.local_rank}_{key}/train', var.item(), self.batch_total)for key, var in speed_stats.items():- self.writer.add_scalar(f'rank{self.local_rank}_{key}/train', eval(var),- epoch * len(self.dataloader_train) + batch_idx)-- # if batch_idx == 2:- # break+ self.writer.add_scalar(f'rank{self.local_rank}_{key}/train', eval(var), self.batch_total)++pbar.close()def _validate_epoch(self, epoch):@@ -356,7 +353,10 @@ class Trainer:if (batch_idx+1) % self.log_interval == 0 or (batch_idx+1) == len(self.dataloader_val):pbar.update(self.log_interval)+ time_now = datetime.now()+ time_now = time_now.strftime("%Y-%m-%d %H:%M:%S")description = (+ f"{time_now}, "f"rank: {self.local_rank}, "f"validation epoch: {epoch}/{self.max_epoch}, "f"step: {batch_idx+1}/{len(self.dataloader_val)}, "diff --git a/funasr/version.txt b/funasr/version.txtindex 238d6e882..b0f3d96f8 100644--- a/funasr/version.txt+++ b/funasr/version.txt@@ -1 +1 @@-1.0.7+1.0.8