add paraformer online opt infer code

haoneng.lhn

2023-09-13 5f088a67cd1b18a8260746971f32a6569e0cf2c6

add paraformer online opt infer code

 funasr/bin/asr_inference_launch.py

@@ -840,37 +840,72 @@
            data = yaml.load(f, Loader=yaml.Loader)
        return data

    def _prepare_cache(cache: dict = {}, chunk_size=[5, 10, 5], batch_size=1):
    def _prepare_cache(cache: dict = {}, chunk_size=[5, 10, 5], encoder_chunk_look_back=0,
                       decoder_chunk_look_back=0, batch_size=1):
        if len(cache) > 0:
            return cache
        config = _read_yaml(asr_train_config)
        enc_output_size = config["encoder_conf"]["output_size"]
        feats_dims = config["frontend_conf"]["n_mels"] * config["frontend_conf"]["lfr_m"]
        cache_en = {"start_idx": 0, "cif_hidden": torch.zeros((batch_size, 1, enc_output_size)),
                    "cif_alphas": torch.zeros((batch_size, 1)), "chunk_size": chunk_size, "last_chunk": False,
                    "cif_alphas": torch.zeros((batch_size, 1)), "chunk_size": chunk_size,
                    "encoder_chunk_look_back": encoder_chunk_look_back, "last_chunk": False, "opt": None,
                    "feats": torch.zeros((batch_size, chunk_size[0] + chunk_size[2], feats_dims)), "tail_chunk": False}
        cache["encoder"] = cache_en

        cache_de = {"decode_fsmn": None}
        cache_de = {"decode_fsmn": None, "decoder_chunk_look_back": decoder_chunk_look_back, "opt": None}
        cache["decoder"] = cache_de

        return cache

    def _cache_reset(cache: dict = {}, chunk_size=[5, 10, 5], batch_size=1):
    def _cache_reset(cache: dict = {}, chunk_size=[5, 10, 5], encoder_chunk_look_back=0,
                     decoder_chunk_look_back=0, batch_size=1):
        if len(cache) > 0:
            config = _read_yaml(asr_train_config)
            enc_output_size = config["encoder_conf"]["output_size"]
            feats_dims = config["frontend_conf"]["n_mels"] * config["frontend_conf"]["lfr_m"]
            cache_en = {"start_idx": 0, "cif_hidden": torch.zeros((batch_size, 1, enc_output_size)),
                        "cif_alphas": torch.zeros((batch_size, 1)), "chunk_size": chunk_size, "last_chunk": False,
                        "feats": torch.zeros((batch_size, chunk_size[0] + chunk_size[2], feats_dims)),
                        "tail_chunk": False}
                        "cif_alphas": torch.zeros((batch_size, 1)), "chunk_size": chunk_size,
                        "encoder_chunk_look_back": encoder_chunk_look_back, "last_chunk": False, "opt": None,
                        "feats": torch.zeros((batch_size, chunk_size[0] + chunk_size[2], feats_dims)), "tail_chunk": False}
            cache["encoder"] = cache_en

            cache_de = {"decode_fsmn": None}
            cache_de = {"decode_fsmn": None, "decoder_chunk_look_back": decoder_chunk_look_back, "opt": None}
            cache["decoder"] = cache_de

        return cache

    #def _prepare_cache(cache: dict = {}, chunk_size=[5, 10, 5], batch_size=1):
    #    if len(cache) > 0:
    #        return cache
    #    config = _read_yaml(asr_train_config)
    #    enc_output_size = config["encoder_conf"]["output_size"]
    #    feats_dims = config["frontend_conf"]["n_mels"] * config["frontend_conf"]["lfr_m"]
    #    cache_en = {"start_idx": 0, "cif_hidden": torch.zeros((batch_size, 1, enc_output_size)),
    #                "cif_alphas": torch.zeros((batch_size, 1)), "chunk_size": chunk_size, "last_chunk": False,
    #                "feats": torch.zeros((batch_size, chunk_size[0] + chunk_size[2], feats_dims)), "tail_chunk": False}
    #    cache["encoder"] = cache_en

    #    cache_de = {"decode_fsmn": None}
    #    cache["decoder"] = cache_de

    #    return cache

    #def _cache_reset(cache: dict = {}, chunk_size=[5, 10, 5], batch_size=1):
    #    if len(cache) > 0:
    #        config = _read_yaml(asr_train_config)
    #        enc_output_size = config["encoder_conf"]["output_size"]
    #        feats_dims = config["frontend_conf"]["n_mels"] * config["frontend_conf"]["lfr_m"]
    #        cache_en = {"start_idx": 0, "cif_hidden": torch.zeros((batch_size, 1, enc_output_size)),
    #                    "cif_alphas": torch.zeros((batch_size, 1)), "chunk_size": chunk_size, "last_chunk": False,
    #                    "feats": torch.zeros((batch_size, chunk_size[0] + chunk_size[2], feats_dims)),
    #                    "tail_chunk": False}
    #        cache["encoder"] = cache_en

    #        cache_de = {"decode_fsmn": None}
    #        cache["decoder"] = cache_de

    #    return cache

    def _forward(
            data_path_and_name_and_type,
@@ -899,12 +934,20 @@
        is_final = False
        cache = {}
        chunk_size = [5, 10, 5]
        encoder_chunk_look_back = 0
        decoder_chunk_look_back = 0
        if param_dict is not None and "cache" in param_dict:
            cache = param_dict["cache"]
        if param_dict is not None and "is_final" in param_dict:
            is_final = param_dict["is_final"]
        if param_dict is not None and "chunk_size" in param_dict:
            chunk_size = param_dict["chunk_size"]
        if param_dict is not None and "encoder_chunk_look_back" in param_dict:
            encoder_chunk_look_back = param_dict["encoder_chunk_look_back"]
            if encoder_chunk_look_back > 0:
                chunk_size[0] = 0
        if param_dict is not None and "decoder_chunk_look_back" in param_dict:
            decoder_chunk_look_back = param_dict["decoder_chunk_look_back"]

        # 7 .Start for-loop
        # FIXME(kamo): The output format should be discussed about
@@ -916,7 +959,8 @@
            sample_offset = 0
            speech_length = raw_inputs.shape[1]
            stride_size = chunk_size[1] * 960
            cache = _prepare_cache(cache, chunk_size=chunk_size, batch_size=1)
            cache = _prepare_cache(cache, chunk_size=chunk_size, batch_size=1, 
                                   encoder_chunk_look_back=encoder_chunk_look_back, decoder_chunk_look_back=decoder_chunk_look_back)
            final_result = ""
            for sample_offset in range(0, speech_length, min(stride_size, speech_length - sample_offset)):
                if sample_offset + stride_size >= speech_length - 1:
@@ -937,7 +981,8 @@

        asr_result_list.append(item)
        if is_final:
            cache = _cache_reset(cache, chunk_size=chunk_size, batch_size=1)
            cache = _cache_reset(cache, chunk_size=chunk_size, batch_size=1,
                                 encoder_chunk_look_back=encoder_chunk_look_back, decoder_chunk_look_back=decoder_chunk_look_back)
        return asr_result_list

    return _forward

 funasr/models/encoder/sanm_encoder.py

@@ -114,8 +114,44 @@
        if not self.normalize_before:
            x = self.norm2(x)


        return x, mask, cache, mask_shfit_chunk, mask_att_chunk_encoder

    def forward_chunk(self, x, cache=None, chunk_size=None, look_back=0):
        """Compute encoded features.

        Args:
            x_input (torch.Tensor): Input tensor (#batch, time, size).
            mask (torch.Tensor): Mask tensor for the input (#batch, time).
            cache (torch.Tensor): Cache tensor of the input (#batch, time - 1, size).

        Returns:
            torch.Tensor: Output tensor (#batch, time, size).
            torch.Tensor: Mask tensor (#batch, time).

        """

        residual = x
        if self.normalize_before:
            x = self.norm1(x)

        if self.in_size == self.size:
            attn, cache = self.self_attn.forward_chunk(x, cache, chunk_size, look_back)
            x = residual + attn
        else:
            x, cache = self.self_attn.forward_chunk(x, cache, chunk_size, look_back)

        if not self.normalize_before:
            x = self.norm1(x)

        residual = x
        if self.normalize_before:
            x = self.norm2(x)
        x = residual + self.feed_forward(x)
        if not self.normalize_before:
            x = self.norm2(x)

        return x, cache


class SANMEncoder(AbsEncoder):
    """
@@ -837,11 +873,56 @@
        cache["feats"] = overlap_feats[:, -(cache["chunk_size"][0] + cache["chunk_size"][2]):, :]
        return overlap_feats

    #def forward_chunk(self,
    #                  xs_pad: torch.Tensor,
    #                  ilens: torch.Tensor,
    #                  cache: dict = None,
    #                  ctc: CTC = None,
    #                  ):
    #    xs_pad *= self.output_size() ** 0.5
    #    if self.embed is None:
    #        xs_pad = xs_pad
    #    else:
    #        xs_pad = self.embed(xs_pad, cache)
    #    if cache["tail_chunk"]:
    #        xs_pad = to_device(cache["feats"], device=xs_pad.device)
    #    else:
    #        xs_pad = self._add_overlap_chunk(xs_pad, cache)
    #    encoder_outs = self.encoders0(xs_pad, None, None, None, None)
    #    xs_pad, masks = encoder_outs[0], encoder_outs[1]
    #    intermediate_outs = []
    #    if len(self.interctc_layer_idx) == 0:
    #        encoder_outs = self.encoders(xs_pad, None, None, None, None)
    #        xs_pad, masks = encoder_outs[0], encoder_outs[1]
    #    else:
    #        for layer_idx, encoder_layer in enumerate(self.encoders):
    #            encoder_outs = encoder_layer(xs_pad, None, None, None, None)
    #            xs_pad, masks = encoder_outs[0], encoder_outs[1]
    #            if layer_idx + 1 in self.interctc_layer_idx:
    #                encoder_out = xs_pad

    #                # intermediate outputs are also normalized
    #                if self.normalize_before:
    #                    encoder_out = self.after_norm(encoder_out)

    #                intermediate_outs.append((layer_idx + 1, encoder_out))

    #                if self.interctc_use_conditioning:
    #                    ctc_out = ctc.softmax(encoder_out)
    #                    xs_pad = xs_pad + self.conditioning_layer(ctc_out)

    #    if self.normalize_before:
    #        xs_pad = self.after_norm(xs_pad)

    #    if len(intermediate_outs) > 0:
    #        return (xs_pad, intermediate_outs), None, None
    #    return xs_pad, ilens, None


    def forward_chunk(self,
                      xs_pad: torch.Tensor,
                      ilens: torch.Tensor,
                      cache: dict = None,
                      ctc: CTC = None,
                      ):
        xs_pad *= self.output_size() ** 0.5
        if self.embed is None:
@@ -852,34 +933,25 @@
            xs_pad = to_device(cache["feats"], device=xs_pad.device)
        else:
            xs_pad = self._add_overlap_chunk(xs_pad, cache)
        encoder_outs = self.encoders0(xs_pad, None, None, None, None)
        xs_pad, masks = encoder_outs[0], encoder_outs[1]
        intermediate_outs = []
        if len(self.interctc_layer_idx) == 0:
            encoder_outs = self.encoders(xs_pad, None, None, None, None)
            xs_pad, masks = encoder_outs[0], encoder_outs[1]
        if cache["opt"] is None:
            cache_layer_num = len(self.encoders0) + len(self.encoders)
            new_cache = [None] * cache_layer_num
        else:
            for layer_idx, encoder_layer in enumerate(self.encoders):
                encoder_outs = encoder_layer(xs_pad, None, None, None, None)
                xs_pad, masks = encoder_outs[0], encoder_outs[1]
                if layer_idx + 1 in self.interctc_layer_idx:
                    encoder_out = xs_pad
            new_cache = cache["opt"]

                    # intermediate outputs are also normalized
                    if self.normalize_before:
                        encoder_out = self.after_norm(encoder_out)
        for layer_idx, encoder_layer in enumerate(self.encoders0):
            encoder_outs = encoder_layer.forward_chunk(xs_pad, new_cache[layer_idx], cache["chunk_size"], cache["encoder_chunk_look_back"])
            xs_pad, new_cache[0] = encoder_outs[0], encoder_outs[1]

                    intermediate_outs.append((layer_idx + 1, encoder_out))

                    if self.interctc_use_conditioning:
                        ctc_out = ctc.softmax(encoder_out)
                        xs_pad = xs_pad + self.conditioning_layer(ctc_out)
        for layer_idx, encoder_layer in enumerate(self.encoders):
            encoder_outs = encoder_layer.forward_chunk(xs_pad, new_cache[layer_idx+len(self.encoders0)], cache["chunk_size"], cache["encoder_chunk_look_back"])
            xs_pad, new_cache[layer_idx+1] = encoder_outs[0], encoder_outs[1]

        if self.normalize_before:
            xs_pad = self.after_norm(xs_pad)
        if cache["encoder_chunk_look_back"] > 0:
            cache["opt"] = new_cache

        if len(intermediate_outs) > 0:
            return (xs_pad, intermediate_outs), None, None
        return xs_pad, ilens, None

    def gen_tf2torch_map_dict(self):

 funasr/modules/attention.py

@@ -456,6 +456,38 @@
        att_outs = self.forward_attention(v_h, scores, mask, mask_att_chunk_encoder)
        return att_outs + fsmn_memory

    def forward_chunk(self, x, cache=None, chunk_size=None, look_back=0):
        """Compute scaled dot product attention.

        Args:
            query (torch.Tensor): Query tensor (#batch, time1, size).
            key (torch.Tensor): Key tensor (#batch, time2, size).
            value (torch.Tensor): Value tensor (#batch, time2, size).
            mask (torch.Tensor): Mask tensor (#batch, 1, time2) or
                (#batch, time1, time2).

        Returns:
            torch.Tensor: Output tensor (#batch, time1, d_model).

        """
        q_h, k_h, v_h, v = self.forward_qkv(x)
        if chunk_size is not None and look_back > 0:
            if cache is not None:
                k_h = torch.cat((cache["k"], k_h), dim=2)
                v_h = torch.cat((cache["v"], v_h), dim=2)
                cache["k"] = k_h[:, :, -(look_back * chunk_size[1]):, :]
                cache["v"] = v_h[:, :, -(look_back * chunk_size[1]):, :]
            else:
                cache_tmp = {"k": k_h[:, :, -(look_back * chunk_size[1]):, :],
                             "v": v_h[:, :, -(look_back * chunk_size[1]):, :]}
                cache = cache_tmp
        fsmn_memory = self.forward_fsmn(v, None)
        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, None)
        return att_outs + fsmn_memory, cache


class MultiHeadedAttentionSANMwithMask(MultiHeadedAttentionSANM):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)