开发者社区 > ModelScope模型即服务 > 语音 > 正文

Paraformer语音识别-中文-通用-16k-离线-large-长音频版微调失败

大家好,请教一个问题: 使用 “Paraformer语音识别-中文-通用-16k-离线-large-长音频版”进行微调,报如下错误,还请帮忙!! 错误如下:

[31c7ad19e26c] 2023-04-15 15:33:52,255 (abs_task:1638) INFO: [valid] Batch sampler: LengthBatchSampler(N-batch=1051, batch_bins=1000, sort_in_batch=descending, sort_batch=descending) [31c7ad19e26c] 2023-04-15 15:33:52,255 (abs_task:1640) INFO: [valid] mini-batch sizes summary: N-batch=1051, mean=13.6, min=5, max=28 [31c7ad19e26c] 2023-04-15 15:33:52,381 (trainer:283) INFO: 1/50epoch started Traceback (most recent call last): File "finetune.py", line 35, in modelscope_finetune(params) File "finetune.py", line 22, in modelscope_finetune trainer.train() File "/opt/conda/lib/python3.7/site-packages/modelscope/trainers/audio/asr_trainer.py", line 168, in train self.trainer.run() File "/root/FunASR/funasr/tasks/abs_task.py", line 1134, in run cls.main_worker(args) File "/root/FunASR/funasr/tasks/abs_task.py", line 1443, in main_worker distributed_option=distributed_option, File "/root/FunASR/funasr/train/trainer.py", line 298, in run distributed_option=distributed_option, File "/root/FunASR/funasr/train/trainer.py", line 603, in train_one_epoch retval = model(**batch) File "/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "/root/FunASR/funasr/models/e2e_asr_paraformer.py", line 1075, in forward encoder_out, encoder_out_lens, text, text_lengths File "/root/FunASR/funasr/models/e2e_asr_paraformer.py", line 524, in _calc_att_loss ignore_id=self.ignore_id) ValueError: too many values to unpack (expected 4)

展开
收起
1622007339875640 2023-04-15 15:46:18 567 0
1 条回答
写回答
取消 提交回答
  • 问题已修复,请更新funasr: https://github.com/alibaba-damo-academy/FunASR#installation

    2023-04-23 14:36:01
    赞同 展开评论 打赏

包括语音识别、语音合成、语音唤醒、声学设计及信号处理、声纹识别、音频事件检测等多个领域

相关电子书

更多
阿里云总监课第二期——Neural Network Language Model在语音识别中的应用 立即下载
阿里云总监课第二期——Latency Controlled-BLSTM模型在语音识别中的应用 立即下载
智能语音交互:阿里巴巴的研究与实践 立即下载