faster-whisper：基于 CTranslate2 的高速语音识别｜部署与使用全指南

# WSL2 Ubuntu / Linux 云服务器
python3 -m venv .venv
source .venv/bin/activate

# Windows PowerShell
python -m venv .venv
.\.venv\Scripts\Activate.ps1

pip install faster-whisper

# 安装 master 分支最新代码
pip install --force-reinstall "faster-whisper @ https://github.com/SYSTRAN/faster-whisper/archive/refs/heads/master.tar.gz"

# 安装指定 commit（示例 commit 取自 README）
pip install --force-reinstall "faster-whisper @ https://github.com/SYSTRAN/faster-whisper/archive/a4f1cc8f11433e454c3934442b5e1a4ed5e865c3.tar.gz"

# Linux / WSL2（写入 ~/.bashrc 可持久化）
export HF_ENDPOINT=https://hf-mirror.com

# Windows PowerShell
$env:HF_ENDPOINT="https://hf-mirror.com"

from faster_whisper import WhisperModel

model = WhisperModel("/path/to/local/model-dir")

# 确认环境已激活，模型首次运行自动下载；无系统 FFmpeg 依赖
python your_transcribe_script.py

from faster_whisper import WhisperModel

model_size = "large-v3"

# Run on GPU with FP16
model = WhisperModel(model_size, device="cuda", compute_type="float16")

# or run on GPU with INT8
# model = WhisperModel(model_size, device="cuda", compute_type="int8_float16")
# or run on CPU with INT8
# model = WhisperModel(model_size, device="cpu", compute_type="int8")

segments, info = model.transcribe("audio.mp3", beam_size=5)

print("Detected language '%s' with probability %f" % (info.language, info.language_probability))

for segment in segments:
    print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))

segments, _ = model.transcribe("audio.mp3")
segments = list(segments)  # The transcription will actually run here.

from faster_whisper import WhisperModel, BatchedInferencePipeline

model = WhisperModel("turbo", device="cuda", compute_type="float16")
batched_model = BatchedInferencePipeline(model=model)
segments, info = batched_model.transcribe("audio.mp3", batch_size=16)

for segment in segments:
    print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))

segments, _ = model.transcribe("audio.mp3", word_timestamps=True)

for segment in segments:
    for word in segment.words:
        print("[%.2fs -> %.2fs] %s" % (word.start, word.end, word.word))

segments, _ = model.transcribe(
    "audio.mp3",
    vad_filter=True,
    vad_parameters=dict(min_silence_duration_ms=500),
)

from faster_whisper import WhisperModel

model_size = "distil-large-v3"
model = WhisperModel(model_size, device="cuda", compute_type="float16")
segments, info = model.transcribe(
    "audio.mp3", beam_size=5, language="en", condition_on_previous_text=False
)

# transcribe_zh.py —— 官方 API 组合示例，依赖：pip install faster-whisper
from faster_whisper import WhisperModel

# 1) 初始化模型：3070 8GB 推荐 medium + int8_float16；4090 可改 large-v3 + float16
model = WhisperModel(
    "medium",
    device="cuda",
    compute_type="int8_float16",
    # cpu_threads=4,          # 纯 CPU 时按需设置
    # download_root="./models", # 指定本地模型目录
)

# 2) 转写：固定中文、beam_size 5、开启 VAD、注入领域提示词、开启词级时间戳
segments, info = model.transcribe(
    "koubo.mp3",
    language="zh",                 # 固定中文；自动检测则留 None
    beam_size=5,
    vad_filter=True,               # 过滤静音片段
    vad_parameters=dict(min_silence_duration_ms=500),
    initial_prompt="以下是技术博客普通话口播内容，包含开源、语音合成、深度学习等术语。",
    word_timestamps=True,          # 词级时间戳，便于精细校对字幕
    condition_on_previous_text=True,
)

print("检测语种：%s（概率 %.3f）" % (info.language, info.language_probability))

# 3) 收集段落到列表（generator 必须迭代后才真正转写）
segments = list(segments)

# 4a) 导出纯文本稿
with open("koubo.txt", "w", encoding="utf-8") as f:
    for seg in segments:
        f.write(seg.text.strip() + "\n")

# 4b) 导出 SRT 字幕（段级时间戳）
def fmt_ts(seconds: float) -> str:
    ms = int((seconds % 1) * 1000)
    s = int(seconds)
    h, rem = divmod(s, 3600)
    m, sec = divmod(rem, 60)
    return f"{h:02d}:{m:02d}:{sec:02d},{ms:03d}"

with open("koubo.srt", "w", encoding="utf-8") as f:
    for idx, seg in enumerate(segments, start=1):
        f.write(f"{idx}\n{fmt_ts(seg.start)} --> {fmt_ts(seg.end)}\n{seg.text.strip()}\n\n")

# 5) 控制台预览
for seg in segments:
    print("[%.2fs -> %.2fs] %s" % (seg.start, seg.end, seg.text.strip()))

export HF_ENDPOINT=https://hf-mirror.com   # 国内镜像（WSL2/Linux）
python transcribe_zh.py