Use clap to achieve prompt controlled generation (#223)
* 快速分类音频并把yml格式结果存在训练根目录里 (#190) * Add files via upload * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> * Update models.py * Update webui.py * Update infer.py * Create compress_model.py * 重新提交，更新Gradio推理UI (#193) * Update webui.py * Update webui.py * 更新 train_ms.py * 更新 models.py * 更新 models.py * 更新 models.py * 更新 train_ms.py * 更新 train_ms.py * 更新 models.py * Update preprocess_text.py * Update config.json * Update train_ms.py * Update webui.py (#206) * Add files via upload (#209) * Update train_ms.py * Update train_ms.py * Update preprocess_text.py * Update train_ms.py * fix (#211) * Update emotion_clustering.py * Add files via upload * Update emotion_clustering.py * add cluster center save * Add files via upload * Update config.py * Update default_config.yml * Update config.py * Update config.py * Update emotion_clustering.py * Update emotion_clustering.py * Update config.py * Update emotion_clustering.py * Update emotion_clustering.py * Update webui.py * Update emotion_clustering.py * Update commons.py * Update emotion_clustering.py * Update webui.py * Update webui.py * Add files via upload * Update train_ms.py * Update train_ms.py * Update train_ms.py * Update train_ms.py * Update train_ms.py * Update webui.py * Update emotion_clustering.py * Update emotion_clustering.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix default_config.yml. * Update infer.py * feat: support infer 2.1 models * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix: support infer 2.1 models 兼容bug修复 * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update train_ms.py * Add CLAP * Fix data loader * Fix infer.py * Fix webui.py * Add prompt template * Update clap_gen.py * Fix wrong environ value * Add g for dur disc * Update clap_gen.py * Fix multilang generation * Update config.json * Prompt mode * Improve slice segments performance * Add preprocess webui * Update webui_preprocess.py * Update webui_preprocess.py * Update config.py * Update default_config.yml * Update config.py * Update clap_gen.py * Delete emo_gen.py * Delete get_emo.py * Delete emotional/wav2vec2-large-robust-12-ft-emotion-msp-dim directory * Update README.md * Update README * Split val per lang * Delete emotion_clustering.py * Update default_config.yml * Update default_config.yml * Update config.py * Update preprocess_text.py * Update webui_preprocess.py * Update defalut_config.yml * Update webui_preprocess.py * Update preprocess_text.py * Random augmentation for CLAP * Update data_utils.py * Update preprocess_text.py * Add vq for CLAP features to avoid overfitting * Random dummy inputs * Update webui.py * Update models.py * Update infer.py * Apply Code Formatter Change * Update config.json * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: YYuX-1145 <138500330+YYuX-1145@users.noreply.github.com> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Sora <654163754@qq.com> Co-authored-by: Sihan Wang <wangsihan1995@gmail.com> Co-authored-by: Stardust-minus <Stardust-minus@users.noreply.github.com>
This commit is contained in:
Stardust·减
117
oldVersion/V210/emo_gen.py
Normal file
117
oldVersion/V210/emo_gen.py
Normal file
@@ -0,0 +1,117 @@
|
||||
import librosa
|
||||
import numpy as np
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
from torch.utils.data import Dataset
|
||||
from torch.utils.data import Dataset
|
||||
from transformers import Wav2Vec2Processor
|
||||
from transformers.models.wav2vec2.modeling_wav2vec2 import (
|
||||
Wav2Vec2Model,
|
||||
Wav2Vec2PreTrainedModel,
|
||||
)
|
||||
|
||||
from config import config
|
||||
|
||||
|
||||
class RegressionHead(nn.Module):
|
||||
r"""Classification head."""
|
||||
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
|
||||
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
||||
self.dropout = nn.Dropout(config.final_dropout)
|
||||
self.out_proj = nn.Linear(config.hidden_size, config.num_labels)
|
||||
|
||||
def forward(self, features, **kwargs):
|
||||
x = features
|
||||
x = self.dropout(x)
|
||||
x = self.dense(x)
|
||||
x = torch.tanh(x)
|
||||
x = self.dropout(x)
|
||||
x = self.out_proj(x)
|
||||
|
||||
return x
|
||||
|
||||
|
||||
class EmotionModel(Wav2Vec2PreTrainedModel):
|
||||
r"""Speech emotion classifier."""
|
||||
|
||||
def __init__(self, config):
|
||||
super().__init__(config)
|
||||
|
||||
self.config = config
|
||||
self.wav2vec2 = Wav2Vec2Model(config)
|
||||
self.classifier = RegressionHead(config)
|
||||
self.init_weights()
|
||||
|
||||
def forward(
|
||||
self,
|
||||
input_values,
|
||||
):
|
||||
outputs = self.wav2vec2(input_values)
|
||||
hidden_states = outputs[0]
|
||||
hidden_states = torch.mean(hidden_states, dim=1)
|
||||
logits = self.classifier(hidden_states)
|
||||
|
||||
return hidden_states, logits
|
||||
|
||||
|
||||
class AudioDataset(Dataset):
|
||||
def __init__(self, list_of_wav_files, sr, processor):
|
||||
self.list_of_wav_files = list_of_wav_files
|
||||
self.processor = processor
|
||||
self.sr = sr
|
||||
|
||||
def __len__(self):
|
||||
return len(self.list_of_wav_files)
|
||||
|
||||
def __getitem__(self, idx):
|
||||
wav_file = self.list_of_wav_files[idx]
|
||||
audio_data, _ = librosa.load(wav_file, sr=self.sr)
|
||||
processed_data = self.processor(audio_data, sampling_rate=self.sr)[
|
||||
"input_values"
|
||||
][0]
|
||||
return torch.from_numpy(processed_data)
|
||||
|
||||
|
||||
device = config.emo_gen_config.device
|
||||
model_name = "./emotional/wav2vec2-large-robust-12-ft-emotion-msp-dim"
|
||||
processor = Wav2Vec2Processor.from_pretrained(model_name)
|
||||
model = EmotionModel.from_pretrained(model_name).to(device)
|
||||
|
||||
|
||||
def process_func(
|
||||
x: np.ndarray,
|
||||
sampling_rate: int,
|
||||
model: EmotionModel,
|
||||
processor: Wav2Vec2Processor,
|
||||
device: str,
|
||||
embeddings: bool = False,
|
||||
) -> np.ndarray:
|
||||
r"""Predict emotions or extract embeddings from raw audio signal."""
|
||||
model = model.to(device)
|
||||
y = processor(x, sampling_rate=sampling_rate)
|
||||
y = y["input_values"][0]
|
||||
y = torch.from_numpy(y).unsqueeze(0).to(device)
|
||||
|
||||
# run through model
|
||||
with torch.no_grad():
|
||||
y = model(y)[0 if embeddings else 1]
|
||||
|
||||
# convert to numpy
|
||||
y = y.detach().cpu().numpy()
|
||||
|
||||
return y
|
||||
|
||||
|
||||
def get_emo(path):
|
||||
wav, sr = librosa.load(path, 16000)
|
||||
return process_func(
|
||||
np.expand_dims(wav, 0).astype(np.float64),
|
||||
sr,
|
||||
model,
|
||||
processor,
|
||||
device,
|
||||
embeddings=True,
|
||||
).squeeze(0)
|
||||
Reference in New Issue
Block a user