import openai import gradio as gr import os from pydub import AudioSegment import math import tempfile import time # ✅ 用来加延时 from datetime import datetime # 从环境变量读取 API Key print("程序启动,开始读取 API Key") openai.api_key = os.getenv("OPENAI_API_KEY") print(f"读取到 API Key,长度: {len(openai.api_key) if openai.api_key else 'None'}") # ✅ 新增:检查当前 API 余额 def check_openai_quota(): try: # 通过 billing API 获取配额(需要付费账户) sub = openai.Subscription.retrieve() usage = openai.Usage.retrieve( start_date=datetime.now().strftime("%Y-%m-01"), end_date=datetime.now().strftime("%Y-%m-%d") ) total = sub["hard_limit_usd"] used = usage["total_usage"] / 100 # 转成 USD remain = total - used return remain except Exception as e: print("⚠️ 无法获取配额信息(可能是免费账户)", e) return None def seconds_to_srt_time(seconds: float) -> str: ms = int((seconds - int(seconds)) * 1000) s = int(seconds) % 60 m = (int(seconds) // 60) % 60 h = int(seconds) // 3600 return f"{h:02}:{m:02}:{s:02},{ms:03}" def merge_srt_segments(segments): srt_output = "" counter = 1 for seg in segments: for entry in seg: start = seconds_to_srt_time(entry["start"]) end = seconds_to_srt_time(entry["end"]) text = entry["text"].replace("\n", " ").strip() srt_output += f"{counter}\n{start} --> {end}\n{text}\n\n" counter += 1 return srt_output def transcribe_long_audio_with_timestamps(audio_path, segment_length_minutes=5): try: # ✅ 调用前检查配额 remain = check_openai_quota() if remain is not None and remain <= 0: return "❌ API 配额已用完,请充值或更换 API Key", None segment_length_ms = segment_length_minutes * 60 * 1000 audio = AudioSegment.from_file(audio_path) duration_ms = len(audio) num_segments = math.ceil(duration_ms / segment_length_ms) all_segments_srt = [] for i in range(num_segments): start = i * segment_length_ms end = min((i + 1) * segment_length_ms, duration_ms) segment = audio[start:end] with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as tmp: segment.export(tmp.name, format="mp3") temp_file_path = tmp.name with open(temp_file_path, "rb") as f: response = openai.audio.transcriptions.create( model="whisper-1", file=f, response_format="verbose_json" ) os.remove(temp_file_path) segments = [] for segment_info in response.segments: segments.append({ "start": segment_info.start + start / 1000, "end": segment_info.end + start / 1000, "text": segment_info.text }) all_segments_srt.append(segments) # ✅ 加个延时,避免请求过快 time.sleep(1) srt_text = merge_srt_segments(all_segments_srt) with open("transcript.srt", "w", encoding="utf-8") as f: f.write(srt_text) full_text = "\n\n".join([seg["text"].strip() for segs in all_segments_srt for seg in segs]) return full_text, "transcript.srt" except Exception as e: return f"❌ 出现错误:{str(e)}", None app = gr.Interface( fn=transcribe_long_audio_with_timestamps, inputs=gr.Audio(type="filepath", label="📤 上传音频(自动分段,含时间戳)"), outputs=[ gr.Textbox(label="📝 转录纯文本"), gr.File(label="⬇️ 下载带时间戳字幕 (.srt)") ], title="🎧 Whisper 长音频分段转录(带时间戳 .srt)", description="上传音频,自动分段转录并生成带时间戳的 SRT 字幕文件,时间戳与音频整体时长对应。" ) app.queue() # ✅ 开启队列模式 app.launch() # ✅ 启动应用