前言
Whisper 是一种通用语音识别模型。它是在大量不同音频数据集上进行训练的，也是一个多任务模型，可以执行多语言语音识别、语音翻译和语言识别。
这里呢，我将给出我的一些代码，来帮助你尽快实现【语音转文字】的服务部署。
以下是该AI模块的具体使用方式：
https://github.com/openai/whisper
心得
这是一个不错的语言模型，它支持自动识别语音语种，类似中文、英文、日语等它都能胜任，并且可以实现其他语种转英语翻译的功能，支持附加时间戳的字幕导出功能......
总体来说，它甚至可以与市面上领头的语言识别功能相媲美，并且主要它是开源的。
这是它的一些模型大小、需要的GPU显存、相对执行速度的对应表
这是它在命令行模式下的使用方式，这对想要尝尝鲜的小伙伴们来说，已经够了
tips：
1、首次安装完毕whisper后，执行指令时会给你安装你所选的模型，small、medium等，我的显卡已经不支持我使用medium了
2、关于GPU版本的pytorch，可以参考如下教程（使用CPU版本会比较慢）
https://blog.csdn.net/G541788_/article/details/135437236
python调用
作为一名python从业者，我十分幸运能够读懂一些模块的相关使用，这里我通过修改了一些模块源码调用，实现了在python代码中一键导出语音字幕的功能（这些功能在命令行中已拥有，但是我希望在使用python脚本model方法后再实现该功能，可能这些你并不需要，但随意吧）。
这个模块的cli()方法或许能更好实现这一功能（因为命令行模式，其实就是运行了这个方法，但我根据经验和实际代码来看，这会重复加载model，导致不必要的资源损耗）。
1、__init__.py中加入get_writer，让你能通过whisper模块去使用这个方法
from .transcribe import get_writer2、相关功能代码
import os.path
import whisper
import time
# 这是语种langue参数的解释，或许对你的选择有帮助
LANGUAGES = {
"en": "english",
"zh": "chinese",
"de": "german",
"es": "spanish",
"ru": "russian",
"ko": "korean",
"fr": "french",
"ja": "japanese",
"pt": "portuguese",
"tr": "turkish",
"pl": "polish",
"ca": "catalan",
"nl": "dutch",
"ar": "arabic",
"sv": "swedish",
"it": "italian",
"id": "indonesian",
"hi": "hindi",
"fi": "finnish",
"vi": "vietnamese",
"he": "hebrew",
"uk": "ukrainian",
"el": "greek",
"ms": "malay",
"cs": "czech",
"ro": "romanian",
"da": "danish",
"hu": "hungarian",
"ta": "tamil",
"no": "norwegian",
"th": "thai",
"ur": "urdu",
"hr": "croatian",
"bg": "bulgarian",
"lt": "lithuanian",
"la": "latin",
"mi": "maori",
"ml": "malayalam",
"cy": "welsh",
"sk": "slovak",
"te": "telugu",
"fa": "persian",
"lv": "latvian",
"bn": "bengali",
"sr": "serbian",
"az": "azerbaijani",
"sl": "slovenian",
"kn": "kannada",
"et": "estonian",
"mk": "macedonian",
"br": "breton",
"eu": "basque",
"is": "icelandic",
"hy": "armenian",
"ne": "nepali",
"mn": "mongolian",
"bs": "bosnian",
"kk": "kazakh",
"sq": "albanian",
"sw": "swahili",
"gl": "galician",
"mr": "marathi",
"pa": "punjabi",
"si": "sinhala",
"km": "khmer",
"sn": "shona",
"yo": "yoruba",
"so": "somali",
"af": "afrikaans",
"oc": "occitan",
"ka": "georgian",
"be": "belarusian",
"tg": "tajik",
"sd": "sindhi",
"gu": "gujarati",
"am": "amharic",
"yi": "yiddish",
"lo": "lao",
"uz": "uzbek",
"fo": "faroese",
"ht": "haitian creole",
"ps": "pashto",
"tk": "turkmen",
"nn": "nynorsk",
"mt": "maltese",
"sa": "sanskrit",
"lb": "luxembourgish",
"my": "myanmar",
"bo": "tibetan",
"tl": "tagalog",
"mg": "malagasy",
"as": "assamese",
"tt": "tatar",
"haw": "hawaiian",
"ln": "lingala",
"ha": "hausa",
"ba": "bashkir",
"jw": "javanese",
"su": "sundanese",
"yue": "cantonese",
}
# 以下命令将使用medium模型转录音频文件中的语音：
#
# whisper audio.flac audio.mp3 audio.wav --model medium
# 默认设置（选择模型small）非常适合转录英语。要转录包含非英语语音的音频文件，您可以使用以下选项指定语言--language：
#
# whisper japanese.wav --language Japanese
# 添加--task translate会将演讲翻译成英语：
#
# whisper japanese.wav --language Japanese --task translate
# 其他语言转录为英语
# whisper "E:\voice\恋愛サーキュレーション_(Vocals)_(Vocals).wav" --language ja --task translate
# 这个任务是将audio_files内的声音文件进行字幕导出，以时间戳为单位存储到captions/目录里
audio_files = [r"E:\voice\恋愛サーキュレーション_(Vocals)_(Vocals).wav"]
model = whisper.load_model("small")
output_format = 'all'
writer_args = {
"highlight_words": False,
"max_line_count": None,
"max_line_width": None,
"max_words_per_line": None,
}
for audio_file in audio_files:
now_timestamp = str(int(time.time()))
save_path = f'captions/{now_timestamp}'
if not os.path.exists(save_path):
os.mkdir(save_path)
# language可选
# 中文zh,日语ja,英语en
result = model.transcribe(audio_file, language='ja')
writer = whisper.get_writer(output_format, save_path)
writer(result, audio_file, **writer_args)
print('done: ', audio_file )
