rgj1980298/obsidian_sync_public/asr-transcribe

{
  "asr": {
    "api_base": "https://api.siliconflow.cn/v1",
    "api_key": "sk-xxx",
    "model": "FunAudioLLM/SenseVoiceSmall",
    "language": "zh"
  }
}

# API Key 从 https://cloud.siliconflow.cn/ 获取

curl -X POST "https://api.siliconflow.cn/v1/audio/transcriptions" \
  -H "Authorization: Bearer {API_KEY}" \
  -F "model=FunAudioLLM/SenseVoiceSmall" \
  -F "language=zh" \
  -F "file=@{音频文件路径}"

{"text": "识别出的文字内容..."}

import httpx
with httpx.Client(timeout=600.0) as client:
    with open("audio.wav", "rb") as f:
        resp = client.post(
            "https://api.siliconflow.cn/v1/audio/transcriptions",
            headers={"Authorization": "Bearer {API_KEY}"},
            data={"model": "FunAudioLLM/SenseVoiceSmall", "language": "zh"},
            files={"file": ("audio.wav", f, "audio/wav")},
        )
    text = resp.json()["text"]

curl -X POST "https://api.openai.com/v1/audio/transcriptions" \
  -H "Authorization: Bearer {OPENAI_API_KEY}" \
  -F "model=whisper-1" \
  -F "language=zh" \
  -F "file=@{音频文件路径}"

pip install faster-whisper

from faster_whisper import WhisperModel
model = WhisperModel("large-v3-turbo", device="cuda", compute_type="float16")
segments, info = model.transcribe(
    "audio.wav",
    language="zh",
    initial_prompt="以下是普通话的句子，使用正确的标点符号，包括句号、逗号、问号等。",
)
text = "".join(seg.text for seg in segments)
print(text)

# 自动检测
import ctranslate2
device = "cuda" if ctranslate2.get_cuda_device_count() > 0 else "cpu"
compute_type = "float16" if device == "cuda" else "int8"

import os, httpx
files = [f for f in os.listdir("audio") if f.endswith((".wav", ".mp3"))]
os.makedirs("text", exist_ok=True)
with httpx.Client(timeout=600.0) as client:
    for f in files:
        with open(f"audio/{f}", "rb") as af:
            resp = client.post(
                "https://api.siliconflow.cn/v1/audio/transcriptions",
                headers={"Authorization": "Bearer {API_KEY}"},
                data={"model": "FunAudioLLM/SenseVoiceSmall", "language": "zh"},
                files={"file": (f, af, "audio/wav")},
            )
        text = resp.json()["text"]
        out = f"text/{os.path.splitext(f)[0]}.txt"
        with open(out, "w", encoding="utf-8") as wf:
            wf.write(text)
        print(f"Done: {f} -> {out} ({len(text)} chars)")

export HF_ENDPOINT=https://hf-mirror.com