如何通过AssemblyAI生成逐词同步的SRT字幕文件

要生成逐词拆分且与音频时间轴同步的SRT字幕，你需要直接利用AssemblyAI转录结果中每个单词的时间戳数据，而非使用默认的句子级字幕导出方法。以下是修改后的完整代码：

import assemblyai as aai
import colorama
from colorama import *
import time
import os

aai.settings.api_key = "(cant share)"

print('')
print('')
print(Fore.GREEN + 'Process 3: Creating subtitles...')
print('')
time.sleep(1)
print(Fore.YELLOW + '>> Creating subtitles...')
transcript = aai.Transcriber().transcribe("output/output-tts.mp3")

# 检查转录是否成功
if transcript.status != aai.TranscriptStatus.completed:
    print(Fore.RED + ">> Transcription failed!")
    exit()

# 定义毫秒转SRT时间格式的函数
def ms_to_srt_time(ms):
    seconds = ms // 1000
    milliseconds = ms % 1000
    minutes = seconds // 60
    seconds = seconds % 60
    hours = minutes // 60
    minutes = minutes % 60
    return f"{hours:02d}:{minutes:02d}:{seconds:02d},{milliseconds:03d}"

# 构建逐词SRT内容
subtitles = []
for idx, word in enumerate(transcript.words, start=1):
    start_time = ms_to_srt_time(word.start)
    end_time = ms_to_srt_time(word.end)
    subtitles.append(f"{idx}\n{start_time} --> {end_time}\n{word.text}\n")

# 合并成完整的SRT字符串
subtitles_content = "\n".join(subtitles)

print('>> Created subtitles!')

# 写入文件（用'w'模式覆盖，避免追加旧内容）
with open("subtitles.srt", "w", encoding="utf-8") as f:
    f.write(subtitles_content)

print('')
print('Program >> You are going to have to manually run the last python file [addsubtitles.py] because \n this program needs to close to write down the subtitles')
print('')
time.sleep(7)

关键修改说明 #

• 利用单词时间戳：AssemblyAI返回的transcript.words列表包含每个单词的start（开始时间，毫秒）、end（结束时间，毫秒）和text（单词文本），这是生成逐词字幕的核心数据。
• 时间格式转换：ms_to_srt_time函数将毫秒值转换为SRT标准的HH:MM:SS,mmm格式。
• 字幕条目构建：遍历每个单词生成独立的字幕条目，编号从1开始递增。
• 文件写入优化：使用with语句安全管理文件，并用w模式覆盖写入，避免多次运行时追加旧内容。

修改后生成的SRT示例 #

1
00:00:00,160 --> 00:00:00,320
Put

2
00:00:00,350 --> 00:00:00,410
a

3
00:00:00,440 --> 00:00:00,680
finger

4
00:00:00,710 --> 00:00:00,950
down
...

内容的提问来源于stack exchange，提问作者flip