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
batch
游雁9a8086bdf5c1
diff --git a/funasr/bin/train.py b/funasr/bin/train.pyindex 97516eb8b..d20915c57 100644--- a/funasr/bin/train.py+++ b/funasr/bin/train.py@@ -205,7 +205,6 @@ def main(**kwargs):dataloader_tr, dataloader_val = dataloader.build_iter(epoch, data_split_i=data_split_i, start_step=trainer.start_step)- trainer.start_step = 0trainer.train_epoch(model=model,@@ -218,7 +217,9 @@ def main(**kwargs):writer=writer,data_split_i=data_split_i,data_split_num=dataloader.data_split_num,+ start_step=trainer.start_step,)+ trainer.start_step = 0torch.cuda.empty_cache()diff --git a/funasr/datasets/audio_datasets/update_jsonl.py b/funasr/datasets/audio_datasets/update_jsonl.pyindex ad47e1285..05870fe1f 100644--- a/funasr/datasets/audio_datasets/update_jsonl.py+++ b/funasr/datasets/audio_datasets/update_jsonl.py@@ -50,8 +50,8 @@ def update_data(lines, i):sample_num = len(waveform)source_len = int(sample_num / 16000 * 1000 / 10)source_len_old = data["source_len"]- if (source_len_old - source_len) > 100 or (source_len - source_len_old) > 100:- logging.info(f"old: {source_len_old}, new: {source_len}, wav: {wav_path}")+ # if (source_len_old - source_len) > 100 or (source_len - source_len_old) > 100:+ # logging.info(f"old: {source_len_old}, new: {source_len}, wav: {wav_path}")data["source_len"] = source_lendata["source"] = wav_pathjsonl_line = json.dumps(data, ensure_ascii=False)diff --git a/funasr/train_utils/trainer.py b/funasr/train_utils/trainer.pyindex e86420cdf..a28ca5183 100644--- a/funasr/train_utils/trainer.py+++ b/funasr/train_utils/trainer.py@@ -456,7 +456,7 @@ class Trainer:batch_num_epoch = len(dataloader_train)self.log(epoch,- batch_idx,+ batch_idx + kwargs.get("start_step", 0),step_in_epoch=self.step_in_epoch,batch_num_epoch=batch_num_epoch,lr=lr,