auto frontend

游雁

2024-06-07 4eedd7ad45120598f10b84673401823719dda237

 funasr/datasets/openai_datasets/datasets.py

@@ -51,7 +51,7 @@
        self.batch_size = kwargs.get("batch_size")
        self.batch_type = kwargs.get("batch_type")
        self.prompt_ids_len = 0
        self.retry = kwargs.get("retry", 5)
        self.retry = kwargs.get("retry", 10)

        self.permute = False
        from funasr.frontends.whisper_frontend import WhisperFrontend
@@ -60,6 +60,8 @@
            self.permute = True

        self.pattern = re.compile(r"(<\|startofspeech\|>.*?<\|endofspeech\|>)")
        # self.kwargs = kwargs
        self.max_token_length = kwargs.get("max_token_length", 1024)

    def get_source_len(self, index):
        item = self.index_ds[index]
@@ -77,7 +79,9 @@
        # pdb.set_trace()

        output = None

        for idx in range(self.retry):
            badcase_flag = False
            if idx == 0:
                index_cur = index
            else:
@@ -112,9 +116,14 @@
                            "<|endofspeech|>", ""
                        )
                        if sub_str.startswith("!"):

                            data_src = load_audio_text_image_video(sub_str[1:], fs=self.fs)

                            try:
                                data_src = load_audio_text_image_video(sub_str[1:], fs=self.fs)
                            except Exception as e:
                                logging.error(
                                    f"Loading wav failed! {str(e)}, {traceback.format_exc()}"
                                )
                                badcase_flag = True
                                continue
                            speech, speech_lengths = extract_fbank(
                                data_src,
                                data_type=self.data_type,
@@ -134,6 +143,8 @@
                            source_ids += sub_token
                            fbank_mask_i += [1] * len(sub_token)

                if badcase_flag:
                    continue
                source_mask = [-100] * len(source_ids)
                target_out = f"{target_out}<|im_end|>"
                target_ids = self.tokenizer.encode(target_out)
@@ -142,6 +153,10 @@
                fbank_mask += fbank_mask_i
                fbank_beg.append(fbank_beg_i)

            if len(input_ids) > self.max_token_length:
                badcase_flag = True
            if badcase_flag:
                continue
            input_ids = torch.tensor(input_ids, dtype=torch.int64)
            attention_mask = torch.tensor([len(input_ids)], dtype=torch.int32)
            labels = torch.tensor(labels, dtype=torch.int64)
@@ -193,9 +208,9 @@
        return outputs

    def _filter_badcase(self, outputs, i=0):
        b, t, _ = outputs["speech"].shape
        b, t = outputs["input_ids"].shape

        if b * t > self.batch_size * 1.25:
        if b * t > self.batch_size * 2:
            beg = torch.randint(0, 2, ()).item()
            if b < 2:
                beg = 0
@@ -204,12 +219,12 @@
            )
            for key, data_list in outputs.items():
                outputs[key] = outputs[key][beg : beg + b : 2]

            speech_lengths_max = outputs["speech_lengths"].max().item()
            outputs["speech"] = outputs["speech"][:, :speech_lengths_max, :]
            text_lengths_max = outputs["text_lengths"].max().item()
            outputs["text"] = outputs["text"][:, :text_lengths_max]
            target_mask_lengths_max = outputs["target_mask_lengths"].max().item()
            outputs["target_mask"] = outputs["target_mask"][:, :target_mask_lengths_max]
            #
            # speech_lengths_max = outputs["speech_lengths"].max().item()
            # outputs["speech"] = outputs["speech"][:, :speech_lengths_max, :]
            # text_lengths_max = outputs["text_lengths"].max().item()
            # outputs["text"] = outputs["text"][:, :text_lengths_max]
            # target_mask_lengths_max = outputs["target_mask_lengths"].max().item()
            # outputs["target_mask"] = outputs["target_mask"][:, :target_mask_lengths_max]

        return outputs