update

嘉渊

2023-04-24 6427c834dfd97b1f05c6659cdc7ccf010bf82fe1

 funasr/bin/vad_inference.py

@@ -1,5 +1,6 @@
import argparse
import logging
import os
import sys
import json
from pathlib import Path
@@ -86,7 +87,8 @@

    @torch.no_grad()
    def __call__(
            self, speech: Union[torch.Tensor, np.ndarray], speech_lengths: Union[torch.Tensor, np.ndarray] = None
            self, speech: Union[torch.Tensor, np.ndarray], speech_lengths: Union[torch.Tensor, np.ndarray] = None,
            in_cache: Dict[str, torch.Tensor] = dict()
    ) -> Tuple[List[List[int]], Dict[str, torch.Tensor]]:
        """Inference

@@ -125,11 +127,12 @@
            batch = {
                "feats": feats[:, t_offset:t_offset + step, :],
                "waveform": speech[:, t_offset * 160:min(speech.shape[-1], (t_offset + step - 1) * 160 + 400)],
                "is_final": is_final
                "is_final": is_final,
                "in_cache": in_cache
            }
            # a. To device
            batch = to_device(batch, device=self.device)
            segments_part = self.vad_model(**batch)
            segments_part, in_cache = self.vad_model(**batch)
            if segments_part:
                for batch_num in range(0, self.batch_size):
                    segments[batch_num] += segments_part[batch_num]
@@ -264,7 +267,8 @@
            # do vad segment
            _, results = speech2vadsegment(**batch)
            for i, _ in enumerate(keys):
                results[i] = json.dumps(results[i])
                if "MODELSCOPE_ENVIRONMENT" in os.environ and os.environ["MODELSCOPE_ENVIRONMENT"] == "eas":
                    results[i] = json.dumps(results[i])
                item = {'key': keys[i], 'value': results[i]}
                vad_results.append(item)
                if writer is not None: