| from .speech_encoder import WhisperWrappedEncoder, DualWrappedEncoder | |
| import torch.nn as nn | |
| def build_speech_encoder(config): | |
| speech_encoder_type = getattr(config, 'speech_encoder_type', None) | |
| print(f"Building speech encoder: {speech_encoder_type}") | |
| if "whisper" in speech_encoder_type.lower(): | |
| return WhisperWrappedEncoder.load(config) | |
| elif "dual" in speech_encoder_type.lower(): | |
| return DualWrappedEncoder(config) | |
| elif "none" in speech_encoder_type.lower(): | |
| return None | |
| raise ValueError(f'Unknown speech encoder: {speech_encoder_type}') | |