声纹识别能制定加载离线模型路径吗

[lonngxiang]

声纹识别能制定加载离线模型路径吗

[lonngxiang]

能指定下载的离线模型吗

from paddlespeech.cli.vector import VectorExecutor
import paddle 
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=r'C:*36145389.wav',
    device=paddle.get_device())
print('Audio embedding Result: \n{}'.format(audio_emb))

[chenzhen2018]

配置说明，源码里有：

https://github.com/PaddlePaddle/PaddleSpeech/blob/969f33999fd6116a69831d098c1ab6f1e125129a/paddlespeech/cli/vector/infer.py#L317

即使用ckpt_path、config两个参数配置 比如下载的模型文件，如下：

可以根据需要将模型文件移动到指定路径下，如/nfs/models/，代码如下：

• 增加ckpt_path、config两个参数

  
  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=r'C:*36145389.wav',
  device=paddle.get_device()),
  ckpt_path="/nfs/models/sv0_ecapa_tdnn_voxceleb12_ckpt_0_1_2/model/model",
  config="/nfs/models/sv0_ecapa_tdnn_voxceleb12_ckpt_0_1_2/conf/model.yaml"

注意：`ckpt_path`必须为`***/xxx/model"，因为源码中会有进一步的处理
```python
os.path.abspath(ckpt_path + ".pdparams")