python/FunASR-XL.git

			@@ -280,6 +280,7 @@
			nbest: int = 1,
			frontend_conf: dict = None,
			hotword_list_or_file: str = None,
			clas_scale: float = 1.0,
			decoding_ind: int = 0,
			**kwargs,
			):
			@@ -376,6 +377,7 @@
			# 6. [Optional] Build hotword list from str, local file or url
			self.hotword_list = None
			self.hotword_list = self.generate_hotwords_list(hotword_list_or_file)
			self.clas_scale = clas_scale

			is_use_lm = lm_weight != 0.0 and lm_file is not None
			if (ctc_weight == 0.0 or asr_model.ctc == None) and not is_use_lm:
			@@ -397,7 +399,7 @@
			@torch.no_grad()
			def __call__(
			self, speech: Union[torch.Tensor, np.ndarray], speech_lengths: Union[torch.Tensor, np.ndarray] = None,
			begin_time: int = 0, end_time: int = None,
			decoding_ind: int = None, begin_time: int = 0, end_time: int = None,
			):
			"""Inference

			@@ -427,7 +429,9 @@
			batch = to_device(batch, device=self.device)

			# b. Forward Encoder
			enc, enc_len = self.asr_model.encode(**batch, ind=self.decoding_ind)
			if decoding_ind is None:
			decoding_ind = self.decoding_ind
			enc, enc_len = self.asr_model.encode(**batch, ind=decoding_ind)
			if isinstance(enc, tuple):
			enc = enc[0]
			# assert len(enc) == 1, len(enc)
			@@ -439,16 +443,20 @@
			pre_token_length = pre_token_length.round().long()
			if torch.max(pre_token_length) < 1:
			return []
			if not isinstance(self.asr_model, ContextualParaformer) and not isinstance(self.asr_model,
			NeatContextualParaformer):
			if not isinstance(self.asr_model, ContextualParaformer) and \
			not isinstance(self.asr_model, NeatContextualParaformer):
			if self.hotword_list:
			logging.warning("Hotword is given but asr model is not a ContextualParaformer.")
			decoder_outs = self.asr_model.cal_decoder_with_predictor(enc, enc_len, pre_acoustic_embeds,
			pre_token_length)
			decoder_out, ys_pad_lens = decoder_outs[0], decoder_outs[1]
			else:
			decoder_outs = self.asr_model.cal_decoder_with_predictor(enc, enc_len, pre_acoustic_embeds,
			pre_token_length, hw_list=self.hotword_list)
			decoder_outs = self.asr_model.cal_decoder_with_predictor(enc,
			enc_len,
			pre_acoustic_embeds,
			pre_token_length,
			hw_list=self.hotword_list,
			clas_scale=self.clas_scale)
			decoder_out, ys_pad_lens = decoder_outs[0], decoder_outs[1]

			if isinstance(self.asr_model, BiCifParaformer):
			@@ -1330,6 +1338,7 @@
			nbest: int = 1,
			streaming: bool = False,
			simu_streaming: bool = False,
			full_utt: bool = False,
			chunk_size: int = 16,
			left_context: int = 32,
			right_context: int = 0,
			@@ -1424,6 +1433,7 @@
			self.beam_search = beam_search
			self.streaming = streaming
			self.simu_streaming = simu_streaming
			self.full_utt = full_utt
			self.chunk_size = max(chunk_size, 0)
			self.left_context = left_context
			self.right_context = max(right_context, 0)
			@@ -1443,6 +1453,7 @@
			self._ctx = self.asr_model.encoder.get_encoder_input_size(
			self.window_size
			)
			self._right_ctx = right_context

			self.last_chunk_length = (
			self.asr_model.encoder.embed.min_frame_length + self.right_context + 1
			@@ -1540,6 +1551,37 @@
			return nbest_hyps

			@torch.no_grad()
			def full_utt_decode(self, speech: Union[torch.Tensor, np.ndarray]) -> List[HypothesisTransducer]:
			"""Speech2Text call.
			Args:
			speech: Speech data. (S)
			Returns:
			nbest_hypothesis: N-best hypothesis.
			"""
			assert check_argument_types()

			if isinstance(speech, np.ndarray):
			speech = torch.tensor(speech)

			if self.frontend is not None:
			speech = torch.unsqueeze(speech, axis=0)
			speech_lengths = speech.new_full([1], dtype=torch.long, fill_value=speech.size(1))
			feats, feats_lengths = self.frontend(speech, speech_lengths)
			else:
			feats = speech.unsqueeze(0).to(getattr(torch, self.dtype))
			feats_lengths = feats.new_full([1], dtype=torch.long, fill_value=feats.size(1))

			if self.asr_model.normalize is not None:
			feats, feats_lengths = self.asr_model.normalize(feats, feats_lengths)

			feats = to_device(feats, device=self.device)
			feats_lengths = to_device(feats_lengths, device=self.device)
			enc_out = self.asr_model.encoder.full_utt_forward(feats, feats_lengths)
			nbest_hyps = self.beam_search(enc_out[0])

			return nbest_hyps

			@torch.no_grad()
			def __call__(self, speech: Union[torch.Tensor, np.ndarray]) -> List[HypothesisTransducer]:
			"""Speech2Text call.
			Args: