EmphAssess 重音检测 · 913 条 · 2026-11-10
同一份权重,两种问法。问答格式下模型大多答错;把 <PCAP> 接在音频后面并预填半句话,让它自己写出重读的词,F1 从 22 升到 79。
| 方式 | system | P | R | F1 | 副口径 F1 | 无答案 |
|---|---|---|---|---|---|---|
| 基线:问答格式 | 带 | 19.70 | 24.92 | 22.01 | 26.34 | 10 |
| 预填:直接取引号里的词 | 带 | 79.29 | 79.37 | 79.33 | 86.42 | 0 |
And so we went straight into, like, feeding baby lambs.
标准答案:重读 we。下面的 <音频> 代表这条语音的一串 CV3 音频 code。
<|im_start|>system You are a helpful assistant. Always follow the user's instructions precisely and respond with what is required. Please read the instructions carefully before answering. Your responses must strictly align with the user's intent and constraints. <|im_end|> <|im_start|>user And so we went straight into, like, feeding baby lambs.[SEP_1]<音频>[SEP_2]<|im_end|> <PCAP>Based on the audio prosody, the speaker stresses the word "we" and speaks with a high level of energy, making the delivery sound excited and lively.</PCAP>
取引号里的词当答案:we对
这一行没有问句,也没有 assistant 回答那一步,答案直接从 <PCAP> 里读。
<|im_start|>system (同一段 system 提示)<|im_end|> <|im_start|>user Answer the question from the user. Put your answer in the format "Emphasized word(s): "___" ". Do not repeat your answer. Audio and prosody: And so we went straight into, like, feeding baby lambs.[SEP_1]<音频>[SEP_2] Which word(s) in the audio clip are emphasized?<|im_end|> <|im_start|>assistant Emphasized word(s): feeding
回答:feeding错
回答之后模型还接着生成了一段语音,检测只对文字部分打分。