Funasr1.0 (#1284)

zhifu gao

2024-01-23 2c3183b61148c622a063edf686440673667c2ce2

Funasr1.0 (#1284)

* funasr1.0 update

* funasr1.0 paraformer-en

* update with main (#1281)

* Funasr1.0 (#1279)

* funasr1.0 update

* funasr1.0 paraformer-en

* update speaker infer

* update device

* update device

* update raw_text

* update infer

* update

* update infer

* bug fix

---------

Co-authored-by: shixian.shi <shixian.shi@alibaba-inc.com>
Co-authored-by: Shi Xian <40013335+R1ckShi@users.noreply.github.com>

* punc bugfix

* reduce_channels

---------

Co-authored-by: shixian.shi <shixian.shi@alibaba-inc.com>
Co-authored-by: Shi Xian <40013335+R1ckShi@users.noreply.github.com>

 funasr/models/ct_transformer/model.py

@@ -333,12 +333,13 @@
                elif new_mini_sentence[-1] == ",":
                    new_mini_sentence_out = new_mini_sentence[:-1] + "."
                    new_mini_sentence_punc_out = new_mini_sentence_punc[:-1] + [self.sentence_end_id]
                elif new_mini_sentence[-1] != "。" and new_mini_sentence[-1] != "？" and len(new_mini_sentence[-1].encode())==0:
                elif new_mini_sentence[-1] != "。" and new_mini_sentence[-1] != "？" and len(new_mini_sentence[-1].encode())!=1:
                    new_mini_sentence_out = new_mini_sentence + "。"
                    new_mini_sentence_punc_out = new_mini_sentence_punc[:-1] + [self.sentence_end_id]
                elif new_mini_sentence[-1] != "." and new_mini_sentence[-1] != "?" and len(new_mini_sentence[-1].encode())==1:
                    new_mini_sentence_out = new_mini_sentence + "."
                    new_mini_sentence_punc_out = new_mini_sentence_punc[:-1] + [self.sentence_end_id]
                    
            # keep a punctuations array for punc segment
            if punc_array is None:
                punc_array = punctuations

 funasr/utils/load_utils.py

@@ -39,7 +39,8 @@
    if isinstance(data_or_path_or_list, str) and os.path.exists(data_or_path_or_list): # local file
        if data_type is None or data_type == "sound":
            data_or_path_or_list, audio_fs = torchaudio.load(data_or_path_or_list)
            data_or_path_or_list = data_or_path_or_list[0, :]
            if kwargs.get("reduce_channels", True):
                data_or_path_or_list = data_or_path_or_list.mean(0)
        elif data_type == "text" and tokenizer is not None:
            data_or_path_or_list = tokenizer.encode(data_or_path_or_list)
        elif data_type == "image": # undo