🚀 Akashpb13/Kabyle_xlsr
该模型是基于MOZILLA - FOUNDATION/COMMON_VOICE_7_0 - hu数据集，对[facebook/wav2vec2 - xls - r - 300m](https://huggingface.co/facebook/wav2vec2 - xls - r - 300m)进行微调后的版本。它能有效处理语音识别相关任务，在语音识别评估中展现出较好的性能。
✨ 主要特性
- 基于预训练模型“facebook/wav2vec2 - xls - r - 300m”进行微调，能更好地适应特定语音数据集。
- 在评估集上取得了较好的结果，如Loss为0.159032，Wer为0.187934 。
📚 详细文档
模型描述
对“facebook/wav2vec2 - xls - r - 300m”进行了微调。
预期用途与限制
更多信息待补充。
训练和评估数据
- 训练数据：Common voice Kabyle train.tsv。由于数据集规模巨大，仅随机采样了50,000条记录进行训练。只考虑那些点赞数大于反对数的点，并且在合并Common Voice 7.0中所有给定数据集后去除了重复项。
训练过程
为创建训练数据集，将所有可能的数据集进行了合并，并采用了90 - 10的分割方式。
训练超参数
训练过程中使用了以下超参数：
| 属性 |
详情 |
| 学习率 |
0.000096 |
| 训练批次大小 |
8 |
| 随机种子 |
13 |
| 梯度累积步数 |
4 |
| 学习率调度器类型 |
cosine_with_restarts |
| 学习率调度器热身步数 |
500 |
| 训练轮数 |
30 |
| 混合精度训练 |
Native AMP |
训练结果
| 步数 |
训练损失 |
验证损失 |
字错误率(Wer) |
| 500 |
7.199800 |
3.130564 |
1.000000 |
| 1000 |
1.570200 |
0.718097 |
0.734682 |
| 1500 |
0.850800 |
0.524227 |
0.640532 |
| 2000 |
0.712200 |
0.468694 |
0.603454 |
| 2500 |
0.651200 |
0.413833 |
0.573025 |
| 3000 |
0.603100 |
0.403680 |
0.552847 |
| 3500 |
0.553300 |
0.372638 |
0.541719 |
| 4000 |
0.537200 |
0.353759 |
0.531191 |
| 4500 |
0.506300 |
0.359109 |
0.519601 |
| 5000 |
0.479600 |
0.343937 |
0.511336 |
| 5500 |
0.479800 |
0.338214 |
0.503948 |
| 6000 |
0.449500 |
0.332600 |
0.495221 |
| 6500 |
0.439200 |
0.323905 |
0.492635 |
| 7000 |
0.434900 |
0.310417 |
0.484555 |
| 7500 |
0.403200 |
0.311247 |
0.483262 |
| 8000 |
0.401500 |
0.295637 |
0.476566 |
| 8500 |
0.397000 |
0.301321 |
0.471672 |
| 9000 |
0.371600 |
0.295639 |
0.468440 |
| 9500 |
0.370700 |
0.294039 |
0.468902 |
| 10000 |
0.364900 |
0.291195 |
0.468440 |
| 10500 |
0.348300 |
0.284898 |
0.461098 |
| 11000 |
0.350100 |
0.281764 |
0.459805 |
| 11500 |
0.336900 |
0.291022 |
0.461606 |
| 12000 |
0.330700 |
0.280467 |
0.455234 |
| 12500 |
0.322500 |
0.271714 |
0.452694 |
| 13000 |
0.307400 |
0.289519 |
0.455465 |
| 13500 |
0.309300 |
0.281922 |
0.451217 |
| 14000 |
0.304800 |
0.271514 |
0.452186 |
| 14500 |
0.288100 |
0.286801 |
0.446830 |
| 15000 |
0.293200 |
0.276309 |
0.445399 |
| 15500 |
0.289800 |
0.287188 |
0.446230 |
| 16000 |
0.274800 |
0.286406 |
0.441243 |
| 16500 |
0.271700 |
0.284754 |
0.441520 |
| 17000 |
0.262500 |
0.275431 |
0.442167 |
| 17500 |
0.255500 |
0.276575 |
0.439858 |
| 18000 |
0.260200 |
0.269911 |
0.435425 |
| 18500 |
0.250600 |
0.270519 |
0.434686 |
| 19000 |
0.243300 |
0.267655 |
0.437826 |
| 19500 |
0.240600 |
0.277109 |
0.431731 |
| 20000 |
0.237200 |
0.266622 |
0.433994 |
| 20500 |
0.231300 |
0.273015 |
0.428868 |
| 21000 |
0.227200 |
0.263024 |
0.430161 |
| 21500 |
0.220400 |
0.272880 |
0.429607 |
| 22000 |
0.218600 |
0.272340 |
0.426883 |
| 22500 |
0.213100 |
0.277066 |
0.428407 |
| 23000 |
0.205000 |
0.278404 |
0.424020 |
| 23500 |
0.200900 |
0.270877 |
0.418987 |
| 24000 |
0.199000 |
0.289120 |
0.425821 |
| 24500 |
0.196100 |
0.275831 |
0.424066 |
| 25000 |
0.191100 |
0.282822 |
0.421850 |
| 25500 |
0.190100 |
0.275820 |
0.418248 |
| 26000 |
0.178800 |
0.279208 |
0.419125 |
| 26500 |
0.183100 |
0.271464 |
0.419218 |
| 27000 |
0.177400 |
0.280869 |
0.419680 |
| 27500 |
0.171800 |
0.279593 |
0.414924 |
| 28000 |
0.172900 |
0.276949 |
0.417648 |
| 28500 |
0.164900 |
0.283491 |
0.417786 |
| 29000 |
0.164800 |
0.283122 |
0.416078 |
| 29500 |
0.165500 |
0.281969 |
0.415801 |
| 30000 |
0.163800 |
0.283319 |
0.412753 |
| 30500 |
0.153500 |
0.285702 |
0.414046 |
| 31000 |
0.156500 |
0.285041 |
0.412615 |
| 31500 |
0.150900 |
0.284336 |
0.413723 |
| 32000 |
0.151800 |
0.285922 |
0.412292 |
| 32500 |
0.149200 |
0.289461 |
0.412153 |
| 33000 |
0.145400 |
0.291322 |
0.409567 |
| 33500 |
0.145600 |
0.294361 |
0.409614 |
| 34000 |
0.144200 |
0.290686 |
0.409059 |
| 34500 |
0.143400 |
0.289474 |
0.409844 |
| 35000 |
0.143500 |
0.290340 |
0.408367 |
| 35500 |
0.143200 |
0.289581 |
0.407351 |
| 36000 |
0.138400 |
0.292782 |
0.408736 |
| 36500 |
0.137900 |
0.289108 |
0.408044 |
| 37000 |
0.138200 |
0.292127 |
0.407166 |
| 37500 |
0.134600 |
0.291797 |
0.408413 |
| 38000 |
0.139800 |
0.290056 |
0.408090 |
| 38500 |
0.136500 |
0.291198 |
0.408090 |
| 39000 |
0.137700 |
0.289696 |
0.408044 |
框架版本
- Transformers 4.16.0.dev0
- Pytorch 1.10.0+cu102
- Datasets 1.18.3
- Tokenizers 0.10.3
评估命令
- 在
mozilla - foundation/common_voice_8_0的test分割上进行评估：
python eval.py --model_id Akashpb13/Kabyle_xlsr --dataset mozilla - foundation/common_voice_8_0 --config kab --split test
📄 许可证
本项目采用Apache - 2.0许可证。
%20--%3e%3cdefs%3e%3cstyle%3e%20.st0%20{%20fill:%20%23061b40;%20}%20.st1%20{%20fill:%20%23306af1;%20}%20.st2%20{%20fill:%20%235ce5cf;%20}%20%3c/style%3e%3c/defs%3e%3cg%3e%3cpath%20class='st0'%20d='M55,10.5h9v3h-9v9h12V7.5h-12v3ZM64,19.5h-6v-3h6v3Z'/%3e%3cpolygon%20class='st0'%20points='69%2016.5%2078%2016.5%2078%2019.5%2069%2019.5%2069%2022.5%2081%2022.5%2081%2013.5%2072%2013.5%2072%2010.5%2081%2010.5%2081%207.5%2069%207.5%2069%2016.5'/%3e%3cpolygon%20class='st0'%20points='95%2010.5%2095%207.5%2083%207.5%2083%2022.5%2095%2022.5%2095%2019.5%2086%2019.5%2086%2016.5%2095%2016.5%2095%2013.5%2086%2013.5%2086%2010.5%2095%2010.5'/%3e%3cpath%20class='st0'%20d='M40,1.5v21h11.6l1.4-1.4v-7.6h0c0,0-1.4-1.5-1.4-1.5l1.4-1.4V2.9l-1.4-1.4h-11.6ZM50,19.5h-7v-6h7v6ZM50,10.5h-7v-6h7v6Z'/%3e%3c/g%3e%3cpath%20class='st1'%20d='M23.1,24L14.7,4.8l-4.9,11.2h3.8l-1.8,4H3.3L12.1,0H2C.9,0,0,.9,0,2v20c0,1.1.9,2,2,2h21.1Z'/%3e%3cpath%20class='st2'%20d='M34,0h-16.8l10.6,24h6.2c1.1,0,2-.9,2-2V2C36,.9,35.1,0,34,0ZM32.5,20h-4V4h4v16Z'/%3e%3c/svg%3e)
Transformers Supports Multiple Languages