将语音识别结果words字段数据转换为DataFrame的Python代码

给定语音识别结果字典result，其segments数组内的每个元素都包含words字段，该字段是一组字典构成的列表，每个字典包含text、start、end等键。需要编写Python代码，提取所有words里的text、start、end数据并转换为Pandas DataFrame。

result的结构示例如下：

result={
    "text": " A nation is a community of people formed on the basis of a combination of shared features. A nation is thus the collective identity of a group of people understood as defined by those features.",
    "segments": [
        {
            "id": 0,
            "seek": 0,
            "start": 0.54,
            "end": 5.08,
            "text": " A nation is a community of people formed on the basis of a combination of shared features.",
            "tokens": [50364,13858,],
            "temperature": 0.0,
            "avg_logprob": -0.3795953684839709,
            "compression_ratio": 1.9613899613899615,
            "no_speech_prob": 0.0013710602652281523,
            "confidence": 0.607,
            "words": [
                {"text": "a", "start": 0.54, "end": 1.3, "confidence": 0.613},
                {"text": "nation", "start": 1.3, "end": 1.98, "confidence": 0.585},
                {"text": "is", "start": 1.98, "end": 2.54, "confidence": 0.974},
                {"text": "community", "start": 2.54, "end": 3.04, "confidence": 0.808},
                {"text": "of", "start": 3.04, "end": 3.48, "confidence": 0.807},
                {"text": "people", "start": 3.48, "end": 3.86, "confidence": 0.764},
                {"text": "form", "start": 3.86, "end": 4.24, "confidence": 0.477},
                {"text": "on.", "start": 4.24, "end": 5.08, "confidence": 0.29}
            ]
        },
        {
            "id": 1,
            "seek": 0,
            "start": 5.38,
            "end": 9.72,
            "text": " the basis of a combination of shared features",
            "tokens": [50614,4704],
            "temperature": 0.0,
            "avg_logprob": -0.3795953684839709,
            "compression_ratio": 1.9613899613899615,
            "no_speech_prob": 0.0013710602652281523,
            "confidence": 0.865,
            "words": [
                {"text": "the", "start": 5.38, "end": 5.8, "confidence": 0.813},
                {"text": "basis", "start": 5.8, "end": 6.2, "confidence": 0.886},
                {"text": "of", "start": 6.2, "end": 6.54, "confidence": 0.961},
                {"text": "a", "start": 6.54, "end": 7.14, "confidence": 0.612},
                {"text": "combination", "start": 7.14, "end": 7.8, "confidence": 0.887},
                {"text": "of", "start": 7.8, "end": 8.18, "confidence": 0.971},
                {"text": "shared", "start": 8.18, "end": 8.5, "confidence": 0.895},
                {"text": "features", "start": 8.5, "end": 8.72, "confidence": 0.88}
            ]
        },
        {"id": 2,...}
    ],
    "language": "ko"
}

通过遍历segments数组，逐个提取每个分段里的words列表，再从每个单词字典中筛选目标字段，最后用Pandas转换为DataFrame即可。

代码实现 #

import pandas as pd

# 初始化空列表存储单词数据
word_data = []

# 遍历所有分段
for segment in result["segments"]:
    # 遍历当前分段下的每个单词
    for word in segment["words"]:
        # 提取需要的字段并加入列表
        word_data.append({
            "text": word["text"],
            "start": word["start"],
            "end": word["end"]
        })

# 转换为DataFrame
df = pd.DataFrame(word_data)

# 可选：打印查看结果
print(df)

代码说明 #

1. 导入pandas库，这是处理结构化表格数据的核心工具。
2. 创建空列表word_data，用来统一收集所有符合要求的单词信息。
3. 嵌套遍历：先遍历所有语音分段，再遍历每个分段下的单词字典。
4. 从每个单词字典中提取text、start、end三个字段，组装成新字典后加入列表。
5. 用pd.DataFrame()将列表转换为结构化的DataFrame，方便后续分析或导出。

扩展优化 #

如果需要保留confidence等其他字段，只需在添加字典时补充对应键值对即可，示例如下：

word_data.append({
    "text": word["text"],
    "start": word["start"],
    "end": word["end"],
    "confidence": word["confidence"]
})

内容的提问来源于stack exchange，提问作者user20189397