通过阿里语音识别调用抱脸空间填写密钥音频转文字
完整 Python 示例
import os
import re
import tempfile
from pathlib import Path
import gradio as gr
from gradio_client import Client, handle_file
# pip install gradio gradio_client
# https://huggingface.co/settings/tokens
SPACE_ID = "Qwen/Qwen3-ASR"
LANGUAGES = [
"Auto",
"Chinese",
"Cantonese",
"English",
"Japanese",
"Korean",
"Arabic",
"German",
"French",
"Spanish",
"Portuguese",
"Indonesian",
"Italian",
"Russian",
"Thai",
"Vietnamese",
"Turkish",
"Hindi",
"Malay",
"Dutch",
"Swedish",
"Danish",
"Finnish",
"Polish",
"Czech",
"Filipino",
"Persian",
"Greek",
"Romanian",
"Hungarian",
"Macedonian",
]
def srt_time(seconds):
seconds = max(0, float(seconds))
ms = int(round(seconds * 1000))
h = ms // 3600000
ms %= 3600000
m = ms // 60000
ms %= 60000
s = ms // 1000
ms %= 1000
return f"{h:02}:{m:02}:{s:02},{ms:03}"
def is_cjk(text):
return bool(
re.search(
r"[\u3400-\u9fff"
r"\u3040-\u30ff"
r"\uac00-\ud7af]",
text,
)
)
def merge_text(old, new):
old = old or ""
new = new or ""
if not old:
return new
if not new:
return old
# 中文、日文、韩文通常不插入空格
if is_cjk(old[-1:]) or is_cjk(new[:1]):
return old + new
# 标点前不加空格
if re.match(r"^[,.;:!?，。！？；：、…]", new):
return old + new
# 撇号等英文情况
if new.startswith(("'", "’")):
return old + new
return old + " " + new
def timestamps_to_segments(
timestamps,
max_chars=28,
max_duration=6.0,
):
"""
将词/字级时间戳合并成适合 SRT 的字幕段。
"""
if not timestamps:
return []
segments = []
current_text = ""
start = None
end = None
punctuation = (
"。！？!?"
"；;"
"\n"
)
for item in timestamps:
text = str(item.get("text", "")).strip()
if not text:
continue
ts_start = float(item.get("start_time", 0) or 0)
ts_end = float(item.get("end_time", ts_start) or ts_start)
if start is None:
start = ts_start
end = ts_end
current_text = merge_text(current_text, text)
duration = end - start
should_split = False
if len(current_text) >= max_chars:
should_split = True
if duration >= max_duration:
should_split = True
if current_text.endswith(punctuation):
should_split = True
if should_split:
segments.append(
{
"start": start,
"end": end,
"text": current_text.strip(),
}
)
current_text = ""
start = None
end = None
if current_text and start is not None:
segments.append(
{
"start": start,
"end": end,
"text": current_text.strip(),
}
)
return segments
def segments_to_srt(segments):
result = []
for i, seg in enumerate(segments, 1):
result.append(str(i))
result.append(
f"{srt_time(seg['start'])} --> "
f"{srt_time(seg['end'])}"
)
result.append(seg["text"])
result.append("")
return "\n".join(result)
def recognize(
audio_file,
hf_token,
language,
max_chars,
max_duration,
):
if not audio_file:
raise gr.Error("请先上传音频。")
token = (hf_token or "").strip()
try:
if token:
client = Client(
SPACE_ID,
token=token,
)
else:
client = Client(SPACE_ID)
result = client.predict(
audio_upload=handle_file(audio_file),
lang_disp=language,
return_ts=True,
api_name="/transcribe",
)
except Exception as e:
raise gr.Error(
f"调用 Qwen3-ASR 失败：\n{e}"
)
if not isinstance(result, (list, tuple)):
raise gr.Error(
f"接口返回格式异常：{type(result)}"
)
detected_language = ""
text = ""
timestamps = None
if len(result) >= 1:
detected_language = result[0] or ""
if len(result) >= 2:
text = result[1] or ""
if len(result) >= 3:
timestamps = result[2]
if not timestamps:
raise gr.Error(
"识别成功，但接口没有返回时间戳，"
"因此暂时无法生成准确 SRT。"
)
segments = timestamps_to_segments(
timestamps,
max_chars=int(max_chars),
max_duration=float(max_duration),
)
srt_content = segments_to_srt(segments)
# 输出文件
audio_path = Path(audio_file)
out_dir = Path(tempfile.gettempdir()) / "qwen3_asr_srt"
out_dir.mkdir(
parents=True,
exist_ok=True,
)
srt_path = out_dir / f"{audio_path.stem}.srt"
srt_path.write_text(
srt_content,
encoding="utf-8-sig",
)
return (
detected_language,
text,
srt_content,
str(srt_path),
)
with gr.Blocks(
title="Qwen3-ASR SRT 字幕工具"
) as demo:
gr.Markdown(
"""
# 🎤 Qwen3-ASR 字幕生成器
上传音频，调用 Qwen3-ASR 自动识别并生成 **SRT 字幕**。
- 支持自动语言检测
- 支持中文 / 英文 / 日文等
- 自动生成时间轴
- 可填写 Hugging Face Token
- 本地不运行 ASR 模型，只调用远程 Space
"""
)
with gr.Row():
with gr.Column():
audio = gr.Audio(
label="🎤 上传音频",
type="filepath",
sources=["upload", "microphone"],
)
token = gr.Textbox(
label="🔑 Hugging Face API Key / Token（可选）",
placeholder="hf_xxxxxxxxxxxxxxxxx",
type="password",
)
language = gr.Dropdown(
LANGUAGES,
value="Auto",
label="🌍 语言",
)
with gr.Row():
max_chars = gr.Slider(
10,
80,
value=28,
step=1,
label="每条字幕最大字符数",
)
max_duration = gr.Slider(
1,
15,
value=6,
step=0.5,
label="每条字幕最长秒数",
)
submit = gr.Button(
"🚀 开始识别并生成 SRT",
variant="primary",
)
with gr.Column():
detected = gr.Textbox(
label="🌍 检测语言"
)
text_result = gr.Textbox(
label="📝 完整识别文本",
lines=10,
)
srt_result = gr.Textbox(
label="🎬 SRT 字幕",
lines=18,
)
srt_file = gr.File(
label="⬇️ 下载 SRT"
)
submit.click(
recognize,
inputs=[
audio,
token,
language,
max_chars,
max_duration,
],
outputs=[
detected,
text_result,
srt_result,
srt_file,
],
)
if __name__ == "__main__":
demo.launch(
server_name="127.0.0.1",
server_port=7860,
inbrowser=True,
)