fix export_meta import of sense voice (#2334)

Zhanzhao (Deo) Liang

2024-12-25 8c7b7e5feb68fda1fc4ddd627bad0f915358149e

 funasr/models/sense_voice/model.py

@@ -19,6 +19,7 @@


from funasr.models.paraformer.search import Hypothesis
from .utils.ctc_alignment import ctc_forced_align


class SinusoidalPositionEncoder(torch.nn.Module):
@@ -94,7 +95,7 @@
        n_feat,
        dropout_rate,
        kernel_size,
        sanm_shfit=0,
        sanm_shift=0,
        lora_list=None,
        lora_rank=8,
        lora_alpha=16,
@@ -120,17 +121,17 @@
        )
        # padding
        left_padding = (kernel_size - 1) // 2
        if sanm_shfit > 0:
            left_padding = left_padding + sanm_shfit
        if sanm_shift > 0:
            left_padding = left_padding + sanm_shift
        right_padding = kernel_size - 1 - left_padding
        self.pad_fn = nn.ConstantPad1d((left_padding, right_padding), 0.0)

    def forward_fsmn(self, inputs, mask, mask_shfit_chunk=None):
    def forward_fsmn(self, inputs, mask, mask_shift_chunk=None):
        b, t, d = inputs.size()
        if mask is not None:
            mask = torch.reshape(mask, (b, -1, 1))
            if mask_shfit_chunk is not None:
                mask = mask * mask_shfit_chunk
            if mask_shift_chunk is not None:
                mask = mask * mask_shift_chunk
            inputs = inputs * mask

        x = inputs.transpose(1, 2)
@@ -210,7 +211,7 @@

        return self.linear_out(x)  # (batch, time1, d_model)

    def forward(self, x, mask, mask_shfit_chunk=None, mask_att_chunk_encoder=None):
    def forward(self, x, mask, mask_shift_chunk=None, mask_att_chunk_encoder=None):
        """Compute scaled dot product attention.

        Args:
@@ -225,7 +226,7 @@

        """
        q_h, k_h, v_h, v = self.forward_qkv(x)
        fsmn_memory = self.forward_fsmn(v, mask, mask_shfit_chunk)
        fsmn_memory = self.forward_fsmn(v, mask, mask_shift_chunk)
        q_h = q_h * self.d_k ** (-0.5)
        scores = torch.matmul(q_h, k_h.transpose(-2, -1))
        att_outs = self.forward_attention(v_h, scores, mask, mask_att_chunk_encoder)
@@ -325,7 +326,7 @@
        self.stochastic_depth_rate = stochastic_depth_rate
        self.dropout_rate = dropout_rate

    def forward(self, x, mask, cache=None, mask_shfit_chunk=None, mask_att_chunk_encoder=None):
    def forward(self, x, mask, cache=None, mask_shift_chunk=None, mask_att_chunk_encoder=None):
        """Compute encoded features.

        Args:
@@ -362,7 +363,7 @@
                    self.self_attn(
                        x,
                        mask,
                        mask_shfit_chunk=mask_shfit_chunk,
                        mask_shift_chunk=mask_shift_chunk,
                        mask_att_chunk_encoder=mask_att_chunk_encoder,
                    ),
                ),
@@ -378,7 +379,7 @@
                    self.self_attn(
                        x,
                        mask,
                        mask_shfit_chunk=mask_shfit_chunk,
                        mask_shift_chunk=mask_shift_chunk,
                        mask_att_chunk_encoder=mask_att_chunk_encoder,
                    )
                )
@@ -387,7 +388,7 @@
                    self.self_attn(
                        x,
                        mask,
                        mask_shfit_chunk=mask_shfit_chunk,
                        mask_shift_chunk=mask_shift_chunk,
                        mask_att_chunk_encoder=mask_att_chunk_encoder,
                    )
                )
@@ -401,7 +402,7 @@
        if not self.normalize_before:
            x = self.norm2(x)

        return x, mask, cache, mask_shfit_chunk, mask_att_chunk_encoder
        return x, mask, cache, mask_shift_chunk, mask_att_chunk_encoder

    def forward_chunk(self, x, cache=None, chunk_size=None, look_back=0):
        """Compute encoded features.
@@ -468,7 +469,7 @@
        positionwise_conv_kernel_size: int = 1,
        padding_idx: int = -1,
        kernel_size: int = 11,
        sanm_shfit: int = 0,
        sanm_shift: int = 0,
        selfattention_layer_type: str = "sanm",
        **kwargs,
    ):
@@ -493,7 +494,7 @@
            output_size,
            attention_dropout_rate,
            kernel_size,
            sanm_shfit,
            sanm_shift,
        )
        encoder_selfattn_layer_args = (
            attention_heads,
@@ -501,7 +502,7 @@
            output_size,
            attention_dropout_rate,
            kernel_size,
            sanm_shfit,
            sanm_shift,
        )

        self.encoders0 = nn.ModuleList(
@@ -555,7 +556,8 @@
        ilens: torch.Tensor,
    ):
        """Embed positions in tensor."""
        masks = sequence_mask(ilens, device=ilens.device)[:, None, :]
        maxlen = xs_pad.shape[1]
        masks = sequence_mask(ilens, maxlen=maxlen, device=ilens.device)[:, None, :]

        xs_pad *= self.output_size() ** 0.5

@@ -856,6 +858,8 @@

        use_itn = kwargs.get("use_itn", False)
        textnorm = kwargs.get("text_norm", None)
        output_timestamp = kwargs.get("output_timestamp", False)

        if textnorm is None:
            textnorm = "withitn" if use_itn else "woitn"
        textnorm_query = self.embed(
@@ -904,13 +908,64 @@
            # Change integer-ids to tokens
            text = tokenizer.decode(token_int)

            result_i = {"key": key[i], "text": text}
            results.append(result_i)
            # result_i = {"key": key[i], "text": text}
            # results.append(result_i)

            if ibest_writer is not None:
                ibest_writer["text"][key[i]] = text

            if output_timestamp:
                from itertools import groupby

                timestamp = []
                tokens = tokenizer.text2tokens(text)[4:]
                logits_speech = self.ctc.softmax(encoder_out)[i, 4 : encoder_out_lens[i].item(), :]
                pred = logits_speech.argmax(-1).cpu()
                logits_speech[pred == self.blank_id, self.blank_id] = 0
                align = ctc_forced_align(
                    logits_speech.unsqueeze(0).float(),
                    torch.Tensor(token_int[4:]).unsqueeze(0).long().to(logits_speech.device),
                    (encoder_out_lens - 4).long(),
                    torch.tensor(len(token_int) - 4).unsqueeze(0).long().to(logits_speech.device),
                    ignore_id=self.ignore_id,
                )
                pred = groupby(align[0, : encoder_out_lens[0]])
                _start = 0
                token_id = 0
                ts_max = encoder_out_lens[i] - 4
                for pred_token, pred_frame in pred:
                    _end = _start + len(list(pred_frame))
                    if pred_token != 0:
                        ts_left = max((_start * 60 - 30) / 1000, 0)
                        ts_right = min((_end * 60 - 30) / 1000, (ts_max * 60 - 30) / 1000)
                        timestamp.append([tokens[token_id], ts_left, ts_right])
                        token_id += 1
                    _start = _end
                timestamp = self.post(timestamp)
                result_i = {"key": key[i], "text": text, "timestamp": timestamp}
                results.append(result_i)
            else:
                result_i = {"key": key[i], "text": text}
                results.append(result_i)
        return results, meta_data

    def post(self, timestamp):
        timestamp_new = []
        for i, t in enumerate(timestamp):
            word, start, end = t
            if word == "▁":
                continue
            if i == 0:
                # timestamp_new.append([word, start, end])
                timestamp_new.append([int(start * 1000), int(end * 1000)])
            elif word.startswith("▁") or len(word) == 1 or not word[1].isalpha():
                word = word[1:]
                # timestamp_new.append([word, start, end])
                timestamp_new.append([int(start * 1000), int(end * 1000)])
            else:
                # timestamp_new[-1][0] += word
                timestamp_new[-1][1] = int(end * 1000)
        return timestamp_new

    def export(self, **kwargs):
        from .export_meta import export_rebuild_model
@@ -919,3 +974,5 @@
            kwargs["max_seq_len"] = 512
        models = export_rebuild_model(model=self, **kwargs)
        return models

        return results, meta_data