update

speech_asr

2023-03-14 4d2bf9fe3cc385b441e94c3000b34a44cac8a8db

update

 funasr/models/frontend/wav_frontend.py

@@ -7,6 +7,7 @@
import torch
import torchaudio.compliance.kaldi as kaldi
from funasr.models.frontend.abs_frontend import AbsFrontend
import funasr.models.frontend.eend_ola_feature as eend_ola_feature
from typeguard import check_argument_types
from torch.nn.utils.rnn import pad_sequence

@@ -444,3 +445,53 @@
        self.reserve_waveforms = None
        self.input_cache = None
        self.lfr_splice_cache = []


class WavFrontendMel23(AbsFrontend):
    """Conventional frontend structure for ASR.
    """

    def __init__(
            self,
            fs: int = 16000,
            frame_length: int = 25,
            frame_shift: int = 10,
            lfr_m: int = 1,
            lfr_n: int = 1,
    ):
        assert check_argument_types()
        super().__init__()
        self.fs = fs
        self.frame_length = frame_length
        self.frame_shift = frame_shift
        self.lfr_m = lfr_m
        self.lfr_n = lfr_n

    def output_size(self) -> int:
        return self.n_mels * self.lfr_m

    def forward(
            self,
            input: torch.Tensor,
            input_lengths: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
        batch_size = input.size(0)
        feats = []
        feats_lens = []
        for i in range(batch_size):
            waveform_length = input_lengths[i]
            waveform = input[i][:waveform_length]
            waveform = waveform.unsqueeze(0).numpy()
            mat = eend_ola_feature.stft(waveform, self.frame_length, self.frame_shift)
            mat = eend_ola_feature.transform(mat)
            mat = mat.splice(mat, context_size=self.lfr_m)
            mat = mat[::self.lfr_n]
            mat = torch.from_numpy(mat)
            feat_length = mat.size(0)
            feats.append(mat)
            feats_lens.append(feat_length)

        feats_lens = torch.as_tensor(feats_lens)
        feats_pad = pad_sequence(feats,
                                 batch_first=True,
                                 padding_value=0.0)
        return feats_pad, feats_lens