funasr1.0

游雁

2024-01-15 a035d68e860ea6decdf422c0fc04eda4fc4de397

 funasr/models/seaco_paraformer/model.py

@@ -35,7 +35,7 @@
    @contextmanager
    def autocast(enabled=True):
        yield
from funasr.utils.load_utils import load_audio_and_text_image_video, extract_fbank
from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank
from funasr.utils import postprocess_utils
from funasr.utils.datadir_writer import DatadirWriter

@@ -90,7 +90,7 @@
        seaco_decoder = kwargs.get("seaco_decoder", None)
        if seaco_decoder is not None:
            seaco_decoder_conf = kwargs.get("seaco_decoder_conf")
            seaco_decoder_class = tables.decoder_classes.get(seaco_decoder.lower())
            seaco_decoder_class = tables.decoder_classes.get(seaco_decoder)
            self.seaco_decoder = seaco_decoder_class(
                vocab_size=self.vocab_size,
                encoder_output_size=self.inner_dim,
@@ -327,7 +327,7 @@
        
        # extract fbank feats
        time1 = time.perf_counter()
        audio_sample_list = load_audio_and_text_image_video(data_in, fs=frontend.fs, audio_fs=kwargs.get("fs", 16000))
        audio_sample_list = load_audio_text_image_video(data_in, fs=frontend.fs, audio_fs=kwargs.get("fs", 16000))
        time2 = time.perf_counter()
        meta_data["load_data"] = f"{time2 - time1:0.3f}"
        speech, speech_lengths = extract_fbank(audio_sample_list, data_type=kwargs.get("data_type", "sound"),
@@ -337,7 +337,8 @@
        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"])

        # hotword
        self.hotword_list = self.generate_hotwords_list(kwargs.get("hotword", None), tokenizer=tokenizer, frontend=frontend)