You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
11 KiB
11 KiB
(简体中文|English)
声纹识别
介绍
声纹识别是一项用计算机程序自动提取说话人特征的技术。
这个 demo 是从一个给定音频文件中提取说话人特征，它可以通过使用 PaddleSpeech 的单个命令或 python 中的几行代码来实现。
使用方法
1. 安装
请看安装文档。
你可以从easy medium，hard 三种方式中选择一种方式安装。
2. 准备输入
声纹cli demo 的输入应该是一个 WAV 文件（.wav），并且采样率必须与模型的采样率相同。
可以下载此 demo 的示例音频：
# 该音频的内容是数字串 85236145389
wget -c https://paddlespeech.bj.bcebos.com/vector/audio/85236145389.wav
3. 使用方法
-
命令行 (推荐使用)
paddlespeech vector --task spk --input 85236145389.wav echo -e "demo1 85236145389.wav" > vec.job paddlespeech vector --task spk --input vec.job echo -e "demo2 85236145389.wav \n demo3 85236145389.wav" | paddlespeech vector --task spk paddlespeech vector --task score --input "./85236145389.wav ./123456789.wav" echo -e "demo4 85236145389.wav 85236145389.wav \n demo5 85236145389.wav 123456789.wav" > vec.job paddlespeech vector --task score --input vec.job使用方法：
paddlespeech vector --help参数：
input(必须输入)：用于识别的音频文件。task(必须输入): 用于指定vector处理的具体任务，默认是spk。model：声纹任务的模型，默认值：ecapatdnn_voxceleb12。sample_rate：音频采样率，默认值：16000。config：声纹任务的参数文件，若不设置则使用预训练模型中的默认配置，默认值：None。ckpt_path：模型参数文件，若不设置则下载预训练模型使用，默认值：None。device：执行预测的设备，默认值：当前系统下 paddlepaddle 的默认 device。
输出：
[ -1.3251206 7.8606825 -4.620626 0.3000721 2.2648535 -1.1931441 3.0647137 7.673595 -6.0044727 -12.02426 -1.9496069 3.1269536 1.618838 -7.6383104 -1.2299773 -12.338331 2.1373026 -5.3957124 9.717328 5.6752305 3.7805123 3.0597172 3.429692 8.97601 13.174125 -0.53132284 8.9424715 4.46511 -4.4262476 -9.726503 8.399328 7.2239175 -7.435854 2.9441683 -4.3430395 -13.886965 -1.6346735 -10.9027405 -5.311245 3.8007221 3.8976038 -2.1230774 -2.3521194 4.151031 -7.4048667 0.13911647 2.4626107 4.9664545 0.9897574 5.4839754 -3.3574002 10.1340065 -0.6120171 -10.403095 4.6007543 16.00935 -7.7836914 -4.1945305 -6.9368606 1.1789556 11.490801 4.2380238 9.550931 8.375046 7.5089145 -0.65707296 -0.30051577 2.8406055 3.0828028 0.730817 6.148354 0.13766119 -13.424735 -7.7461405 -2.3227983 -8.305252 2.9879124 -10.995229 0.15211068 -2.3820348 -1.7984174 8.495629 -5.8522367 -3.755498 0.6989711 -5.2702994 -2.6188622 -1.8828466 -4.64665 14.078544 -0.5495333 10.579158 -3.2160501 9.349004 -4.381078 -11.675817 -2.8630207 4.5721755 2.246612 -4.574342 1.8610188 2.3767874 5.6257877 -9.784078 0.64967257 -1.4579505 0.4263264 -4.9211264 -2.454784 3.4869802 -0.42654222 8.341269 1.356552 7.0966883 -13.102829 8.016734 -7.1159344 1.8699781 0.208721 14.699384 -1.025278 -2.6107233 -2.5082312 8.427193 6.9138527 -6.2912464 0.6157366 2.489688 -3.4668267 9.921763 11.200815 -0.1966403 7.4916005 -0.62312716 -0.25848144 -9.947997 -0.9611041 1.1649219 -2.1907122 -1.5028487 -0.51926106 15.165954 2.4649463 -0.9980445 7.4416637 -2.0768049 3.5896823 -7.3055434 -7.5620847 4.323335 0.0804418 -6.56401 -2.3148053 -1.7642345 -2.4708817 -7.675618 -9.548878 -1.0177554 0.16986446 2.5877135 -1.8752296 -0.36614323 -6.0493784 -2.3965611 -5.9453387 0.9424033 -13.155974 -7.457801 0.14658108 -3.742797 5.8414927 -1.2872906 5.5694313 12.57059 1.0939219 2.2142086 1.9181576 6.9914207 -5.888139 3.1409824 -2.003628 2.4434285 9.973139 5.03668 2.0051203 2.8615603 5.860224 2.9176188 -1.6311141 2.0292206 -4.070415 -6.831437 ] -
Python API
import paddle from paddlespeech.cli import VectorExecutor vector_executor = VectorExecutor() audio_emb = vector_executor( model='ecapatdnn_voxceleb12', sample_rate=16000, config=None, # Set `config` and `ckpt_path` to None to use pretrained model. ckpt_path=None, audio_file='./85236145389.wav', device=paddle.get_device()) print('Audio embedding Result: \n{}'.format(audio_emb)) test_emb = vector_executor( model='ecapatdnn_voxceleb12', sample_rate=16000, config=None, # Set `config` and `ckpt_path` to None to use pretrained model. ckpt_path=None, audio_file='./123456789.wav', device=paddle.get_device()) print('Test embedding Result: \n{}'.format(test_emb)) # score range [0, 1] score = vector_executor.get_embeddings_score(audio_emb, test_emb) print(f"Eembeddings Score: {score}")输出：
# Vector Result: Audio embedding Result: [ 1.4217498 5.626253 -5.342073 1.1773866 3.308055 1.756596 5.167894 10.80636 -3.8226728 -5.6141334 2.623845 -0.8072968 1.9635103 -7.3128724 0.01103897 -9.723131 0.6619743 -6.976803 10.213478 7.494748 2.9105635 3.8949256 3.7999806 7.1061673 16.905321 -7.1493764 8.733103 3.4230042 -4.831653 -11.403367 11.232214 7.1274667 -4.2828417 2.452362 -5.130748 -18.177666 -2.6116815 -11.000337 -6.7314315 1.6564683 0.7618269 1.1253023 -2.083836 4.725744 -8.782597 -3.539873 3.814236 5.1420674 2.162061 4.096431 -6.4162116 12.747448 1.9429878 -15.152943 6.417416 16.097002 -9.716668 -1.9920526 -3.3649497 -1.871939 11.567354 3.69788 11.258265 7.442363 9.183411 4.5281515 -1.2417862 4.3959084 6.6727695 5.8898783 7.627124 -0.66919386 -11.889693 -9.208865 -7.4274073 -3.7776625 6.917234 -9.848748 -2.0944717 -5.135116 0.49563864 9.317534 -5.9141874 -1.8098574 -0.11738578 -7.169265 -1.0578263 -5.7216787 -5.1173844 16.137651 -4.473626 7.6624317 -0.55381083 9.631587 -6.4704556 -8.548508 4.3716145 -0.79702514 4.478997 -2.9758704 3.272176 2.8382776 5.134597 -9.190781 -0.5657382 -4.8745747 2.3165567 -5.984303 -2.1798875 0.35541576 -0.31784213 9.493548 2.1144536 4.358092 -12.089823 8.451689 -7.925461 4.6242585 4.4289427 18.692003 -2.6204622 -5.149185 -0.35821092 8.488551 4.981496 -9.32683 -2.2544234 6.6417594 1.2119585 10.977129 16.555033 3.3238444 9.551863 -1.6676947 -0.79539716 -8.605674 -0.47356385 2.6741948 -5.359179 -2.6673796 0.66607 15.443222 4.740594 -3.4725387 11.592567 -2.054497 1.7361217 -8.265324 -9.30447 5.4068313 -1.5180256 -7.746615 -6.089606 0.07112726 -0.34904733 -8.649895 -9.998958 -2.564841 -0.53999114 2.601808 -0.31927416 -1.8815292 -2.07215 -3.4105783 -8.2998085 1.483641 -15.365992 -8.288208 3.8847756 -3.4876456 7.3629923 0.4657332 3.132599 12.438889 -1.8337058 4.532936 2.7264361 10.145339 -6.521951 2.897153 -3.3925855 5.079156 7.759716 4.677565 5.8457737 2.402413 7.7071047 3.9711342 -6.390043 6.1268735 -3.7760346 -11.118123 ] # get the test embedding Test embedding Result: [ 2.5247195 5.119042 -4.335273 4.4583654 5.047907 3.5059214 1.6159848 0.49364898 -11.6899185 -3.1014526 -5.6589785 -0.42684984 2.674276 -11.937654 6.2248464 -10.776924 -5.694543 1.112041 1.5709964 1.0961034 1.3976512 2.324352 1.339981 5.279319 13.734659 -2.5753925 13.651442 -2.2357535 5.1575427 -3.251567 1.4023279 6.1191974 -6.0845175 -1.3646189 -2.6789894 -15.220778 9.779349 -9.411551 -6.388947 6.8313975 -9.245996 0.31196198 2.5509644 -4.413065 6.1649427 6.793837 2.6328635 8.620976 3.4832475 0.52491665 2.9115407 5.8392377 0.6702376 -3.2726715 2.6694255 16.91701 -5.5811176 0.23362345 -4.5573606 -11.801059 14.728292 -0.5198082 -3.999922 7.0927105 -7.0459595 -5.4389 -0.46420583 -5.1085467 10.376568 -8.889225 -0.37705845 -1.659806 2.6731026 -7.1909504 1.4608804 -2.163136 -0.17949677 4.0241547 0.11319201 0.601279 2.039692 3.1910992 -11.649526 -8.121584 -4.8707457 0.3851982 1.4231744 -2.3321972 0.99332285 14.121717 5.899413 0.7384519 -17.760096 10.555021 4.1366534 -0.3391071 -0.20792882 3.208204 0.8847948 -8.721497 -6.432868 13.006379 4.8956 -9.155822 -1.9441519 5.7815638 -2.066733 10.425042 -0.8802383 -2.4314315 -9.869258 0.35095334 -5.3549943 2.1076174 -8.290468 8.4433365 -4.689333 9.334139 -2.172678 -3.0250976 8.394216 -3.2110903 -7.93868 2.3960824 -2.3213403 -1.4963245 -3.476059 4.132903 -10.893354 4.362673 -0.45456508 10.258634 -1.1655927 -6.7799754 0.22885278 -4.399287 2.333433 -4.84745 -4.2752337 -1.3577863 -1.0685898 9.505196 7.3062205 0.08708266 12.927811 -9.57974 1.3936648 -1.9444873 5.776769 15.251903 10.6118355 -1.4903594 -9.535318 -3.6553776 -1.6699586 -0.5933151 7.600357 -4.8815503 -8.698617 -15.855757 0.25632986 -7.2235737 0.9506656 0.7128582 -9.051738 8.74869 -1.6426028 -6.5762258 2.506905 -6.7431564 5.129912 -12.189555 -3.6435068 12.068113 -6.0059533 -2.3535995 2.9014351 22.3082 -1.5563312 13.193291 2.7583609 -7.468798 1.3407065 -4.599617 -6.2345777 10.7689295 7.137627 5.099476 0.3473359 9.647881 -2.0484571 -5.8549366 ] # get the score between enroll and test Eembeddings Score: 0.45332613587379456
4.预训练模型
以下是 PaddleSpeech 提供的可以被命令行和 python API 使用的预训练模型列表：
| 模型 | 采样率 |
|---|---|
| ecapatdnn_voxceleb12 | 16k |