wzq15896 commited on
Commit
d0a9d39
·
verified ·
1 Parent(s): b5db286

Upload 3 files

Browse files
Files changed (3) hide show
  1. README.md +192 -15
  2. app.py +368 -0
  3. requirements.txt +15 -0
README.md CHANGED
@@ -1,15 +1,192 @@
1
- ---
2
- title: Develop
3
- emoji: 🏢
4
- colorFrom: purple
5
- colorTo: red
6
- sdk: gradio
7
- sdk_version: 6.19.0
8
- python_version: '3.13'
9
- app_file: app.py
10
- pinned: false
11
- license: mit
12
- short_description: demo3
13
- ---
14
-
15
- Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ title: SpeechBrain Demo
3
+ emoji: 🧠
4
+ colorFrom: blue
5
+ colorTo: indigo
6
+ sdk: gradio
7
+ sdk_version: "4.44.0"
8
+ python_version: "3.10"
9
+ app_file: app.py
10
+ pinned: false
11
+ ---
12
+
13
+ # 🧠 SpeechBrain - 全能语音处理工具包
14
+
15
+ > **SpeechBrain** 是一个基于 PyTorch 的开源语音处理工具包,支持 **语音识别 (ASR)**、**语音合成 (TTS)**、**说话人识别**、**语音增强**、**语音分离**、**语音活动检测 (VAD)** 等多种任务。
16
+
17
+ [![Python](https://img.shields.io/badge/Python-3.8%2B-blue.svg)](https://www.python.org/)
18
+ [![PyTorch](https://img.shields.io/badge/PyTorch-2.1%2B-red.svg)](https://pytorch.org/)
19
+ [![License](https://img.shields.io/badge/License-Apache%202.0-green.svg)](LICENSE)
20
+ [![Version](https://img.shields.io/badge/Version-1.1.0-orange.svg)]()
21
+
22
+ ---
23
+
24
+ ## ✨ 主要功能
25
+
26
+ | 功能 | 描述 | 预训练模型 |
27
+ |------|------|------------|
28
+ | 🎙️ **语音识别 (ASR)** | 将语音转为文字,支持多种语言 | ✅ |
29
+ | 🔊 **语音合成 (TTS)** | 将文字转为自然语音 | ✅ |
30
+ | 👤 **说话人识别** | 识别说话人身份 | ✅ |
31
+ | 🎵 **语音增强** | 降噪、去混响 | ✅ |
32
+ | 🎧 **语音分离** | 分离混合音频中的多个声源 | ✅ |
33
+ | 📊 **语音活动检测** | 检测语音段/非语音段 | ✅ |
34
+ | 📝 **说话人日志** | 谁在什么时候说话 | ✅ |
35
+ | 🌐 **翻译 (ST)** | 语音到文本翻译 | ✅ |
36
+
37
+ ---
38
+
39
+ ## 🚀 快速开始
40
+
41
+ ### 安装
42
+
43
+ ```bash
44
+ # 克隆仓库
45
+ git clone https://github.com/zq-1115/demo.git
46
+ cd demo
47
+
48
+ # 安装依赖
49
+ pip install -r requirements.txt
50
+ ```
51
+
52
+ ### 30秒体验
53
+
54
+ ```python
55
+ from speechbrain.inference import EncoderDecoderASR
56
+
57
+ # 下载预训练模型并转录音频
58
+ asr_model = EncoderDecoderASR.from_hparams(
59
+ source="speechbrain/asr-crdnn-rnnlm-librispeech",
60
+ savedir="pretrained_models/asr-crdnn-rnnlm-librispeech",
61
+ )
62
+
63
+ # 转录音频文件
64
+ text = asr_model.transcribe_file("demo_audio.wav")
65
+ print(f"识别结果: {text}")
66
+ ```
67
+
68
+ ---
69
+
70
+ ## 📂 项目结构
71
+
72
+ ```
73
+ speechbrain-develop/
74
+ ├── speechbrain/ # 核心库代码
75
+ │ ├── inference/ # 推理接口 (ASR/TTS/VAD等)
76
+ │ ├── dataio/ # 数据加载与处理
77
+ │ ├── decoders/ # 解码器 (CTC/Seq2Seq等)
78
+ │ ├── augment/ # 数据增强
79
+ │ ├── alignment/ # 对齐工具
80
+ │ ├── integrations/ # 集成模块 (HuggingFace/K2等)
81
+ │ └── utils/ # 工具函数
82
+ ├── tests/ # 测试与示例音频
83
+ ├── demo.py # 🎯 交互式演示脚本
84
+ ├── requirements.txt # 依赖列表
85
+ └── setup.py # 安装配置
86
+ ```
87
+
88
+ ---
89
+
90
+ ## 🎮 运行演示
91
+
92
+ ```bash
93
+ python demo.py
94
+ ```
95
+
96
+ 演示程序提供交互式菜单,可以选择:
97
+ 1. **语音识别 (ASR)** — 上传音频,输出文字
98
+ 2. **说话人识别** — 识别说话人
99
+ 3. **语音活动检测 (VAD)** — 检测语音片段
100
+
101
+ ---
102
+
103
+ ## 🔧 使用示例
104
+
105
+ ### 1. 语音识别 (ASR)
106
+
107
+ ```python
108
+ from speechbrain.inference import EncoderDecoderASR
109
+
110
+ asr = EncoderDecoderASR.from_hparams(
111
+ source="speechbrain/asr-crdnn-rnnlm-librispeech",
112
+ savedir="pretrained_models/asr",
113
+ )
114
+ result = asr.transcribe_file("audio.wav")
115
+ print(result) # "MY FATHER HAS REVEALED THE CULPRIT'S NAME"
116
+ ```
117
+
118
+ ### 2. 说话人识别
119
+
120
+ ```python
121
+ from speechbrain.inference import SpeakerRecognition
122
+
123
+ spk_rec = SpeakerRecognition.from_hparams(
124
+ source="speechbrain/spkrec-ecapa-voxceleb",
125
+ savedir="pretrained_models/spkrec",
126
+ )
127
+ score, prediction = spk_rec.verify_files("spk1.wav", "spk2.wav")
128
+ print(f"是同一个人吗? {'是' if prediction else '否'}")
129
+ ```
130
+
131
+ ### 3. 语音活动检测 (VAD)
132
+
133
+ ```python
134
+ from speechbrain.inference import VAD
135
+
136
+ vad = VAD.from_hparams(
137
+ source="speechbrain/vad-crdnn-libriparty",
138
+ savedir="pretrained_models/vad",
139
+ )
140
+ boundaries = vad.get_speech_segments("audio.wav")
141
+ for start, end in boundaries:
142
+ print(f"语音段: {start:.2f}s - {end:.2f}s")
143
+ ```
144
+
145
+ ### 4. 语音合成 (TTS)
146
+
147
+ ```python
148
+ from speechbrain.inference import Tacotron2
149
+ from speechbrain.inference import HIFIGAN
150
+
151
+ tacotron2 = Tacotron2.from_hparams(
152
+ source="speechbrain/tts-tacotron2-ljspeech",
153
+ savedir="pretrained_models/tts-tacotron2",
154
+ )
155
+ hifi_gan = HIFIGAN.from_hparams(
156
+ source="speechbrain/tts-hifigan-ljspeech",
157
+ savedir="pretrained_models/tts-hifigan",
158
+ )
159
+
160
+ # 文字转语音
161
+ mel_output, mel_length, alignment = tacotron2.encode_text("Hello world")
162
+ waveforms = hifi_gan.decode_batch(mel_output)
163
+ torchaudio.save("output.wav", waveforms.squeeze(1), 22050)
164
+ ```
165
+
166
+ ---
167
+
168
+ ## 📦 依赖
169
+
170
+ - **Python** >= 3.8
171
+ - **PyTorch** >= 2.1.0
172
+ - **torchaudio** >= 2.1.0
173
+ - **transformers** >= 4.30.0
174
+ - 完整列表见 [requirements.txt](requirements.txt)
175
+
176
+ ---
177
+
178
+ ## 🤝 贡献
179
+
180
+ 欢迎提交 Issue 和 Pull Request!
181
+
182
+ ---
183
+
184
+ ## 📄 许可证
185
+
186
+ 本项目基于 Apache 2.0 许可证开源。详见 [LICENSE](LICENSE)。
187
+
188
+ ---
189
+
190
+ ## ⭐ Star History
191
+
192
+ 如果这个项目对你有帮助,请给一个 ⭐ Star 支持一下!
app.py ADDED
@@ -0,0 +1,368 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ SpeechBrain Hugging Face Spaces Demo
3
+ =====================================
4
+ 基于 Gradio 的交互式演示,用于 Hugging Face Spaces 部署。
5
+
6
+ 功能:
7
+ 1. 语音识别 (ASR) — 将语音转录为文字
8
+ 2. 说话人验证 — 判断两段音频是否为同一人
9
+ 3. 语音活动检测 (VAD) — 检测音频中的语音片段
10
+ 4. 说话人嵌入提取 — 提取声纹特征向量
11
+ """
12
+
13
+ import os
14
+ import sys
15
+ import warnings
16
+
17
+ warnings.filterwarnings("ignore")
18
+
19
+ # ============================================================
20
+ # Gradio / HuggingFace Spaces 兼容性修复
21
+ # ============================================================
22
+
23
+ # --- 修复 1: gradio_client 的 boolean schema 问题 ---
24
+ # gradio_client 在处理 additionalProperties: true 时,
25
+ # 将 boolean True 作为 schema 传入,导致 "const" in schema 崩溃
26
+ try:
27
+ import gradio_client.utils as _gc_utils
28
+
29
+ _original_get_type = _gc_utils.get_type
30
+
31
+ def _patched_get_type(schema):
32
+ if isinstance(schema, bool):
33
+ return "boolean"
34
+ return _original_get_type(schema)
35
+
36
+ _gc_utils.get_type = _patched_get_type
37
+
38
+ _original_json_schema = _gc_utils._json_schema_to_python_type
39
+
40
+ def _patched_json_schema(schema, defs):
41
+ if isinstance(schema, bool):
42
+ return "boolean"
43
+ return _original_json_schema(schema, defs)
44
+
45
+ _gc_utils._json_schema_to_python_type = _patched_json_schema
46
+ except ImportError:
47
+ pass
48
+
49
+ # --- 修复 2: Jinja2 缓存 key 不可哈希问题 ---
50
+ # 某些 gradio 版本传给 Jinja2 的 globals 包含不可哈希 dict
51
+ try:
52
+ import jinja2.environment as _jinja_env
53
+
54
+ _original_load_template = _jinja_env.Environment._load_template
55
+
56
+ def _patched_load_template(self, name, globals):
57
+ try:
58
+ return _original_load_template(self, name, globals)
59
+ except TypeError:
60
+ # 缓存 key 不可哈希时,传入空 dict(而非 None),
61
+ # 因为 Jinja2 会将 None 替换为 self.globals,而后者也可能包含不可哈希类型
62
+ return _original_load_template(self, name, {})
63
+
64
+ _jinja_env.Environment._load_template = _patched_load_template
65
+ except ImportError:
66
+ pass
67
+
68
+ # ============================================================
69
+ # Windows 兼容性修复
70
+ # ============================================================
71
+ if sys.platform == "win32":
72
+ import types
73
+ for _dep in ["k2", "flair", "numba", "spacy"]:
74
+ if _dep not in sys.modules:
75
+ m = types.ModuleType(_dep)
76
+ m.__path__ = []
77
+ sys.modules[_dep] = m
78
+
79
+ import torch
80
+ import gradio as gr
81
+
82
+ # ============================================================
83
+ # 模型缓存目录(HuggingFace Spaces 会自动缓存到 /data 或默认目录)
84
+ # ============================================================
85
+ MODEL_DIR = os.path.join(os.path.dirname(__file__), "pretrained_models")
86
+
87
+ _asr_model = None
88
+ _spk_model = None
89
+ _vad_model = None
90
+
91
+
92
+ def get_device():
93
+ """获取可用设备"""
94
+ return "cuda" if torch.cuda.is_available() else "cpu"
95
+
96
+
97
+ # ============================================================
98
+ # 模型加载(懒加载,首次使用时才下载)
99
+ # ============================================================
100
+
101
+ def get_asr_model():
102
+ """语音识别模型 (CRDNN + RNNLM)"""
103
+ global _asr_model
104
+ if _asr_model is None:
105
+ from speechbrain.inference.ASR import EncoderDecoderASR
106
+ print("[INFO] 正在加载 ASR 模型...")
107
+ _asr_model = EncoderDecoderASR.from_hparams(
108
+ source="speechbrain/asr-crdnn-rnnlm-librispeech",
109
+ savedir=os.path.join(MODEL_DIR, "asr-crdnn-rnnlm-librispeech"),
110
+ run_opts={"device": get_device()},
111
+ )
112
+ return _asr_model
113
+
114
+
115
+ def get_speaker_model():
116
+ """说话人识别模型 (ECAPA-TDNN)"""
117
+ global _spk_model
118
+ if _spk_model is None:
119
+ from speechbrain.inference.speaker import SpeakerRecognition
120
+ print("[INFO] 正在加载说话人识别模型...")
121
+ _spk_model = SpeakerRecognition.from_hparams(
122
+ source="speechbrain/spkrec-ecapa-voxceleb",
123
+ savedir=os.path.join(MODEL_DIR, "spkrec-ecapa-voxceleb"),
124
+ run_opts={"device": get_device()},
125
+ )
126
+ return _spk_model
127
+
128
+
129
+ def get_vad_model():
130
+ """语音活动检测模型 (CRDNN)"""
131
+ global _vad_model
132
+ if _vad_model is None:
133
+ from speechbrain.inference.VAD import VAD
134
+ print("[INFO] 正在加载 VAD 模型...")
135
+ _vad_model = VAD.from_hparams(
136
+ source="speechbrain/vad-crdnn-libriparty",
137
+ savedir=os.path.join(MODEL_DIR, "vad-crdnn-libriparty"),
138
+ run_opts={"device": get_device()},
139
+ )
140
+ return _vad_model
141
+
142
+
143
+ # ============================================================
144
+ # 功能函数
145
+ # ============================================================
146
+
147
+ def transcribe_audio(audio_file):
148
+ """语音识别:将音频转为文字"""
149
+ if audio_file is None:
150
+ return "⚠️ 请先上传音频文件"
151
+
152
+ try:
153
+ asr = get_asr_model()
154
+ text = asr.transcribe_file(audio_file)
155
+ return f"📝 **识别结果:**\n\n> {text}"
156
+ except Exception as e:
157
+ return f"❌ 识别失败:{str(e)}"
158
+
159
+
160
+ def verify_speakers(audio1, audio2):
161
+ """说话人验证:判断两段音频是否为同一说话人"""
162
+ if audio1 is None or audio2 is None:
163
+ return "⚠️ 请上传两段音频文件"
164
+
165
+ try:
166
+ spk = get_speaker_model()
167
+ score, prediction = spk.verify_files(audio1, audio2)
168
+
169
+ similarity = score.item()
170
+ is_same = "✅ 是同一说话人" if prediction else "❌ 不是同一说话人"
171
+
172
+ if similarity > 0.5:
173
+ level = "🟢 高"
174
+ elif similarity > 0.0:
175
+ level = "🟡 中"
176
+ else:
177
+ level = "🔴 低"
178
+
179
+ return (
180
+ f"## {is_same}\n\n"
181
+ f"| 指标 | 值 |\n"
182
+ f"|------|----|\n"
183
+ f"| **相似度得分** | {similarity:.4f} |\n"
184
+ f"| **置信程度** | {level} |\n"
185
+ f"| **判定阈值** | 0.0(高于此值判定为同一人) |"
186
+ )
187
+ except Exception as e:
188
+ return f"❌ 验证失败:{str(e)}"
189
+
190
+
191
+ def detect_speech(audio_file):
192
+ """语音活动检测:标记音频中的语音段"""
193
+ if audio_file is None:
194
+ return "⚠️ 请先上传音频文件"
195
+
196
+ try:
197
+ vad = get_vad_model()
198
+ boundaries = vad.get_speech_segments(audio_file)
199
+
200
+ if len(boundaries) == 0:
201
+ return "🔇 未检测到语音"
202
+
203
+ lines = [
204
+ "## 🎯 检测结果\n",
205
+ f"| 片段 | 开始 | 结束 | 时长 |",
206
+ f"|------|------|------|------|",
207
+ ]
208
+
209
+ total_duration = 0
210
+ for i, (start_t, end_t) in enumerate(boundaries, 1):
211
+ duration = end_t - start_t
212
+ total_duration += duration
213
+ lines.append(
214
+ f"| {i} | {start_t:.2f}s | {end_t:.2f}s | {duration:.2f}s |"
215
+ )
216
+
217
+ lines.append(
218
+ f"\n📊 **共 {len(boundaries)} 个语音片段,总时长 {total_duration:.2f}s**"
219
+ )
220
+ return "\n".join(lines)
221
+ except Exception as e:
222
+ return f"❌ 检测失败:{str(e)}"
223
+
224
+
225
+ def extract_speaker_embedding(audio_file):
226
+ """提取说话人嵌入向量(声纹特征)"""
227
+ if audio_file is None:
228
+ return "⚠️ 请先上传音频文件"
229
+
230
+ try:
231
+ spk = get_speaker_model()
232
+ embedding = spk.encode_file(audio_file)
233
+ shape = tuple(embedding.shape)
234
+ return (
235
+ f"✅ **提取成功!**\n\n"
236
+ f"- 嵌入向量维度:`{shape}`\n"
237
+ f"- 前 10 个值:`{embedding[0, :10].tolist()}`\n"
238
+ f"- 用途:说话人聚类、验证、识别等"
239
+ )
240
+ except Exception as e:
241
+ return f"❌ 提取失败:{str(e)}"
242
+
243
+
244
+ # ============================================================
245
+ # Gradio 界面
246
+ # ============================================================
247
+
248
+ def create_demo():
249
+ device = get_device()
250
+ device_name = "GPU 🚀" if device == "cuda" else "CPU ⚡"
251
+
252
+ with gr.Blocks(
253
+ title="SpeechBrain Demo — 语音处理工具箱",
254
+ theme=gr.themes.Soft(),
255
+ css="""
256
+ .output-markdown { font-size: 16px; }
257
+ footer { visibility: hidden; }
258
+ """,
259
+ ) as demo:
260
+ gr.Markdown(
261
+ f"""
262
+ # 🧠 SpeechBrain 语音处理 Demo
263
+ ### 基于 PyTorch 的全能语音处理工具包 | 运行设备:**{device_name}**
264
+
265
+ 上传音频文件,体验语音识别、说话人验证、语音活动检测等功能。
266
+ 首次使用会自动从 HuggingFace 下载预训练模型。
267
+ """
268
+ )
269
+
270
+ with gr.Tabs():
271
+ # ── Tab 1: 语音识别 ──
272
+ with gr.Tab("🎙️ 语音识别 (ASR)"):
273
+ gr.Markdown("上传英文音频,自动转录为文字。支持上传文件或麦克风录音。")
274
+ with gr.Row():
275
+ with gr.Column(scale=1):
276
+ audio_asr = gr.Audio(
277
+ label="上传音频",
278
+ type="filepath",
279
+ sources=["upload", "microphone"],
280
+ )
281
+ btn_asr = gr.Button("🔍 开始识别", variant="primary", size="lg")
282
+ with gr.Column(scale=1):
283
+ output_asr = gr.Markdown(value="等待上传音频...")
284
+
285
+ btn_asr.click(
286
+ fn=transcribe_audio,
287
+ inputs=audio_asr,
288
+ outputs=output_asr,
289
+ )
290
+
291
+ # ── Tab 2: 说话人验证 ──
292
+ with gr.Tab("👤 说话人验证"):
293
+ gr.Markdown("上传两段音频,判断是否为同一个人说话。")
294
+ with gr.Row():
295
+ with gr.Column():
296
+ audio_spk1 = gr.Audio(label="音频 1", type="filepath")
297
+ with gr.Column():
298
+ audio_spk2 = gr.Audio(label="音频 2", type="filepath")
299
+ btn_spk = gr.Button("🔍 开始验证", variant="primary", size="lg")
300
+ output_spk = gr.Markdown(value="等待上传两段音频...")
301
+
302
+ btn_spk.click(
303
+ fn=verify_speakers,
304
+ inputs=[audio_spk1, audio_spk2],
305
+ outputs=output_spk,
306
+ )
307
+
308
+ # ── Tab 3: 语音活动检测 ──
309
+ with gr.Tab("📊 语音检测 (VAD)"):
310
+ gr.Markdown("检测音频中哪些时间段有语音活动。")
311
+ with gr.Row():
312
+ with gr.Column(scale=1):
313
+ audio_vad = gr.Audio(
314
+ label="上传音频",
315
+ type="filepath",
316
+ sources=["upload", "microphone"],
317
+ )
318
+ btn_vad = gr.Button("🔍 开始检测", variant="primary", size="lg")
319
+ with gr.Column(scale=1):
320
+ output_vad = gr.Markdown(value="等待上传音频...")
321
+
322
+ btn_vad.click(
323
+ fn=detect_speech,
324
+ inputs=audio_vad,
325
+ outputs=output_vad,
326
+ )
327
+
328
+ # ── Tab 4: 说话人嵌入 ──
329
+ with gr.Tab("🧬 声纹提取"):
330
+ gr.Markdown("提取说话人的声纹特征向量(Embedding),可用于下游任务。")
331
+ with gr.Row():
332
+ with gr.Column(scale=1):
333
+ audio_emb = gr.Audio(label="上传音频", type="filepath")
334
+ btn_emb = gr.Button("🔍 提取嵌入", variant="primary", size="lg")
335
+ with gr.Column(scale=1):
336
+ output_emb = gr.Markdown(value="等待上传音频...")
337
+
338
+ btn_emb.click(
339
+ fn=extract_speaker_embedding,
340
+ inputs=audio_emb,
341
+ outputs=output_emb,
342
+ )
343
+
344
+ gr.Markdown(
345
+ """
346
+ ---
347
+ ### 📚 相关链接
348
+ - [SpeechBrain GitHub](https://github.com/speechbrain/speechbrain)
349
+ - [HuggingFace 模型库](https://huggingface.co/speechbrain)
350
+ - 模型:ASR (CRDNN+RNNLM) | 说话人识别 (ECAPA-TDNN) | VAD (CRDNN)
351
+ """
352
+ )
353
+
354
+ return demo
355
+
356
+
357
+ # ============================================================
358
+ # HuggingFace Spaces 入口
359
+ # ============================================================
360
+ if __name__ == "__main__":
361
+ demo = create_demo()
362
+ # HuggingFace Spaces 会自动注入环境变量,本地开发时绑定 0.0.0.0:7860
363
+ server_name = os.environ.get("GRADIO_SERVER_NAME", "0.0.0.0")
364
+ server_port = int(os.environ.get("GRADIO_SERVER_PORT", 7860))
365
+ demo.launch(
366
+ server_name=server_name,
367
+ server_port=server_port,
368
+ )
requirements.txt ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ speechbrain
2
+ gradio>=4.44.0,<5.0.0
3
+ huggingface_hub>=0.8.0,<1.0
4
+ hyperpyyaml>=0.0.1
5
+ joblib>=0.14.1
6
+ numpy>=1.17.0
7
+ packaging
8
+ requests>=2.20.0
9
+ scipy>=1.4.1
10
+ sentencepiece>=0.1.91
11
+ soundfile>=0.12.1
12
+ torch>=2.1.0
13
+ torchaudio>=2.1.0
14
+ tqdm>=4.42.0
15
+ transformers>=4.30.0