funasr1.0 streaming demo

游雁

2024-01-12 0143122a4e2ee86cc27ba137b2bb0530577cbf12

 funasr/models/ct_transformer/model.py

@@ -11,6 +11,7 @@
import torch
import torch.nn as nn
from funasr.models.ct_transformer.utils import split_to_mini_sentence, split_words
from funasr.utils.load_utils import load_audio_text_image_video

from funasr.register import tables

@@ -219,10 +220,10 @@
                 **kwargs,
                 ):
        assert len(data_in) == 1
        
        text = load_audio_text_image_video(data_in, data_type=kwargs.get("kwargs", "text"))[0]
        vad_indexes = kwargs.get("vad_indexes", None)
        text = data_in[0]
        text_lengths = data_lengths[0] if data_lengths is not None else None
        # text = data_in[0]
        # text_lengths = data_lengths[0] if data_lengths is not None else None
        split_size = kwargs.get("split_size", 20)
        
        tokens = split_words(text)
@@ -238,6 +239,7 @@
        cache_pop_trigger_limit = 200
        results = []
        meta_data = {}
        punc_array = None
        for mini_sentence_i in range(len(mini_sentences)):
            mini_sentence = mini_sentences[mini_sentence_i]
            mini_sentence_id = mini_sentences_id[mini_sentence_i]
@@ -319,8 +321,13 @@
                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]

        result_i = {"key": key[0], "text": new_mini_sentence_out}
            # keep a punctuations array for punc segment
            if punc_array is None:
                punc_array = punctuations
            else:
                punc_array = torch.cat([punc_array, punctuations], dim=0)
        
        result_i = {"key": key[0], "text": new_mini_sentence_out, "punc_array": punc_array}
        results.append(result_i)
    
        return results, meta_data