bug fix

shixian.shi

2024-01-12 c3c78fc5e790d48b3a2f9da79199320c06108d38

 funasr/models/fsmn_vad/model.py

@@ -12,6 +12,7 @@
from funasr.utils.load_utils import load_audio_text_image_video,extract_fbank
from funasr.utils.datadir_writer import DatadirWriter
from torch.nn.utils.rnn import pad_sequence
from funasr.train_utils.device_funcs import to_device

class VadStateMachine(Enum):
    kVadInStateStartPointNotDetected = 1
@@ -554,7 +555,8 @@
            meta_data[
                "batch_data_time"] = speech_lengths.sum().item() * frontend.frame_shift * frontend.lfr_n / 1000

        speech.to(device=kwargs["device"]), speech_lengths.to(device=kwargs["device"])
        speech = speech.to(device=kwargs["device"])
        speech_lengths = speech_lengths.to(device=kwargs["device"])

        # b. Forward Encoder streaming
        t_offset = 0
@@ -579,7 +581,8 @@
                "cache": cache
            }


            
            batch = to_device(batch, device=kwargs["device"])
            segments_part, cache = self.forward(**batch)
            if segments_part:
                for batch_num in range(0, batch_size):
@@ -593,15 +596,16 @@
        results = []
        for i in range(batch_size):
            
            if "MODELSCOPE_ENVIRONMENT" in os.environ and os.environ["MODELSCOPE_ENVIRONMENT"] == "eas":
                results[i] = json.dumps(results[i])
                
            if ibest_writer is not None:
                ibest_writer["text"][key[i]] = segments[i]

            result_i = {"key": key[i], "value": segments[i]}
            results.append(result_i)
 
            
            if "MODELSCOPE_ENVIRONMENT" in os.environ and os.environ["MODELSCOPE_ENVIRONMENT"] == "eas":
                results[i] = json.dumps(results[i])
                
        return results, meta_data

    def DetectCommonFrames(self) -> int: