shixian.shi
2023-05-11 8c732716d62a2c2bdf2f302a7904f1811aa4301d
update timestamp for batch inference
1个文件已修改
14 ■■■■ 已修改文件
funasr/bin/asr_inference_paraformer_vad_punc.py 14 ●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/bin/asr_inference_paraformer_vad_punc.py
@@ -3,6 +3,7 @@
import json
import argparse
import logging
from re import L
import sys
import time
import os
@@ -635,6 +636,9 @@
                result[0] += restored_data[j][0]
                result[1] += restored_data[j][1]
                result[2] += restored_data[j][2]
                for t in restored_data[j][4]:
                    t[0] += vadsegments[j][0]
                    t[1] += vadsegments[j][0]
                result[4] += restored_data[j][4]
                # result = [result[k]+restored_data[j][k] for k in range(len(result[:-2]))]
@@ -642,16 +646,6 @@
            # result = result_segments[0]
            text, token, token_int = result[0], result[1], result[2]
            time_stamp = None if len(result) < 5 else result[4]
            # post process timestamp
            offset_list = [0]
            offset = 0
            for i in range(1, len(time_stamp)):
                if time_stamp[i][0] < time_stamp[i-1][1]:
                    offset += time_stamp[i-1][1]
                offset_list.append(offset)
            for i in range(1, len(time_stamp)):
                time_stamp[i][0] += offset_list[i]
                time_stamp[i][1] += offset_list[i]
            if use_timestamp and time_stamp is not None:
                postprocessed_result = postprocess_utils.sentence_postprocess(token, time_stamp)