Improve language tag
#1
by
lbourdois
- opened
README.md
CHANGED
|
@@ -1,104 +1,115 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
datasets:
|
| 4 |
-
- KE-Team/SemanticVAD
|
| 5 |
-
language:
|
| 6 |
-
-
|
| 7 |
-
-
|
| 8 |
-
|
| 9 |
-
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
#
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
#
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- KE-Team/SemanticVAD
|
| 5 |
+
language:
|
| 6 |
+
- zho
|
| 7 |
+
- eng
|
| 8 |
+
- fra
|
| 9 |
+
- spa
|
| 10 |
+
- por
|
| 11 |
+
- deu
|
| 12 |
+
- ita
|
| 13 |
+
- rus
|
| 14 |
+
- jpn
|
| 15 |
+
- kor
|
| 16 |
+
- vie
|
| 17 |
+
- tha
|
| 18 |
+
- ara
|
| 19 |
+
base_model:
|
| 20 |
+
- Qwen/Qwen2.5-0.5B-Instruct
|
| 21 |
+
---
|
| 22 |
+
|
| 23 |
+
# 模型简介 🚀
|
| 24 |
+
本模型是基于Qwen2.5-0.5B-Instruct架构微调的语义级语音活动检测(Semantic Voice Activity Detection)模型，用于支持全双工语音对话系统。
|
| 25 |
+
|
| 26 |
+
# 测试集表现 📈
|
| 27 |
+
| 标签 | 准确率 / % |
|
| 28 |
+
| :----- | :---------- |
|
| 29 |
+
| <打断> | 98.07 |
|
| 30 |
+
| <附和> | 98.12 |
|
| 31 |
+
| <完成> | 92.73 |
|
| 32 |
+
| <未完> | 99.91 |
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
# 基础使用 🛠️
|
| 36 |
+
|
| 37 |
+
```python
|
| 38 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 39 |
+
|
| 40 |
+
# 加载模型
|
| 41 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 42 |
+
"KE-Team/KE-SemanticVAD").to('cuda')
|
| 43 |
+
tokenizer = AutoTokenizer.from_pretrained("KE-Team/KE-SemanticVAD")
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
# System Prompt
|
| 47 |
+
AGNET_SPKING_SYS='# Role\n你是人机实时交互的**用户行为分析**模块，你将收到包含部分历史信息的 Human 和 Agent 最新实时对话记录 (Dialog)\n\n# 任务\n当前【Agent正在发言】，在此过程中，你需要基于对话分析 Human 的意图属于 <打断> 还是 <附和>\n\n# 输出\n不要有多余的分析，仅严格输出以下二者之一: <打断> 或 <附和>\n\n# 判断标准\n## <打断> 的情况\nHuman 行为: 试图抢夺话题主导权\n特征包括:\n- 提供新概念/词汇/判断（如命名、定性、对比）\n- 提出问题或异议\n- 引入与当前话题无关的新话题\n\n## <附和> 的情况\nHuman 行为: 赞同 Agent, 期望 Agent 继续说\n特征包括:\n- 使用零内容反馈（嗯/啊/对）\n- 机械重复 Agent 中的原词/同义词\n- 表达简单的确认或同意（如“是的”、“没错”）\n'
|
| 48 |
+
HUMAN_SPKING_SYS = '# Role\n你是人机实时交互的**用户行为分析**模块，你将收到包含部分历史信息的 Human 和 Agent 最新实时对话记录 (Dialog)\n\n# 任务\n当前【Human正在发言】，你需要基于对话判断 Human 是否已经完成发言\n\n# 输出\n严格输出以下二者之一: <完成> 或 <未完>\n\n# 判断标准\n## <完成> 的情况\nHuman 发言语义完整，说话很可能已经结束\n- 发言包含完整命题（如明确提问/请求/结论）\n- 出现结束性标记词（"好了"/"你觉得呢？"）\n\n## <未完> 的情况\nHuman 发言语义不完整，仍然可能继续说话\n- 语句末尾含连接词（"而且"/"不过"/"然后"）\n- 用户发言中夹杂思考词（"呃..."/"嗯..."）\n'
|
| 49 |
+
SYS_MAP = dict(
|
| 50 |
+
agent = AGNET_SPKING_SYS,
|
| 51 |
+
human = HUMAN_SPKING_SYS
|
| 52 |
+
)
|
| 53 |
+
|
| 54 |
+
# Dialog Format
|
| 55 |
+
def dia_format(x):
|
| 56 |
+
cur_spk = x['speaker']
|
| 57 |
+
system = SYS_MAP[cur_spk]
|
| 58 |
+
if cur_spk == 'agent':
|
| 59 |
+
u1 = x['history']['query']
|
| 60 |
+
a1 = x['history']['answer']
|
| 61 |
+
u2 = x['query']
|
| 62 |
+
dialog = f"# Dialog\nHuman[历史]:{u1}\nAgent:[实时]:{a1}\nHuman[实时]:{u2}\n"
|
| 63 |
+
elif cur_spk == 'human':
|
| 64 |
+
if len(x['history']) <= 1:
|
| 65 |
+
u2 = x['query']
|
| 66 |
+
dialog = f"# Dialog\nHuman[实时]:{u2}\n"
|
| 67 |
+
else:
|
| 68 |
+
u1 = x['history']['query']
|
| 69 |
+
a1 = x['history']['answer']
|
| 70 |
+
u2 = x['query']
|
| 71 |
+
dialog = f"# Dialog\nHuman[历史]:{u1}\nAgent:[历史]:{a1}\nHuman[实时]:{u2}\n"
|
| 72 |
+
else:
|
| 73 |
+
raise ValueError('current speaker should in agent or human')
|
| 74 |
+
return [{'role': 'system', 'content':system}, {'role': 'user', 'content': dialog}]
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
# 数据样例
|
| 78 |
+
example = {
|
| 79 |
+
"speaker": "agent",
|
| 80 |
+
"query": "那具体是怎么实现的？比如说，如",
|
| 81 |
+
"history": {
|
| 82 |
+
"query": "怎么把人工智能技术用在虚拟现实开发上呢？",
|
| 83 |
+
"answer": "将人工智能技术应用到虚拟现实开发中，可以通过智能算法来提升用户体验，比如使用机器学习来创建更真实的虚拟角色"
|
| 84 |
+
}
|
| 85 |
+
}
|
| 86 |
+
messages = dia_format(example)
|
| 87 |
+
text = tokenizer.apply_chat_template(
|
| 88 |
+
messages,
|
| 89 |
+
tokenize=False,
|
| 90 |
+
add_generation_prompt=True
|
| 91 |
+
)
|
| 92 |
+
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
| 93 |
+
generated_ids = model.generate(
|
| 94 |
+
**model_inputs,
|
| 95 |
+
max_new_tokens=64
|
| 96 |
+
)
|
| 97 |
+
generated_ids = [
|
| 98 |
+
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
| 99 |
+
]
|
| 100 |
+
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
| 101 |
+
print(f"检测结果: {response}") # -> <打断>
|
| 102 |
+
```
|
| 103 |
+
|
| 104 |
+
# Citation
|
| 105 |
+
|
| 106 |
+
Please cite our Hugging-Face when using our code, data or model.
|
| 107 |
+
```Bibtext
|
| 108 |
+
@misc{KE-SemanticVAD,
|
| 109 |
+
author = {KE-TEAM},
|
| 110 |
+
title = {KE-SemanticVAD},
|
| 111 |
+
year = {2025},
|
| 112 |
+
publisher = {Hugging Face},
|
| 113 |
+
url = {https://huggingface.co/KE-Team/KE-SemanticVAD}
|
| 114 |
+
}
|
| 115 |
+
```
|