Update README.md
Browse files
README.md
CHANGED
|
@@ -7,11 +7,7 @@ pipeline_tag: audio-classification
|
|
| 7 |
---
|
| 8 |
Leveraging the multi-dimensional fine-grained annotations produced by our pipeline, we introduce **FM-Speech**, built upon the frontier Qwen3-Omni (30B MoE) architecture.
|
| 9 |
|
| 10 |
-
|
| 11 |
-
<!-- 请在此处放一张精简的模型输入输出示意图 -->
|
| 12 |
-
<!--<img width="500px" src="src/model_io.png" alt="Model Input Output Diagram"/> -->
|
| 13 |
-
<p><i>Input: Raw Speech   ➔   Output: 14-Dimension Fine-Grained Speech Attributes (Structured JSON)</i></p>
|
| 14 |
-
</div>
|
| 15 |
|
| 16 |
To overcome modality gaps and text-conditioned hallucinations, FM-Speech is trained using a **Progressive Curriculum Fine-Tuning** framework, decoupling complex auditory comprehension into three incremental stages: Warm-up (MCQ/QA) $\rightarrow$ Capability Ramp-up $\rightarrow$ Final Alignment (Full JSON).
|
| 17 |
|
|
|
|
| 7 |
---
|
| 8 |
Leveraging the multi-dimensional fine-grained annotations produced by our pipeline, we introduce **FM-Speech**, built upon the frontier Qwen3-Omni (30B MoE) architecture.
|
| 9 |
|
| 10 |
+
> 🎙️ **Input:** Raw Speech Audio   ➔   📊 **Output:** 14-Dimension Fine-Grained Speech Attributes (Structured JSON)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
To overcome modality gaps and text-conditioned hallucinations, FM-Speech is trained using a **Progressive Curriculum Fine-Tuning** framework, decoupling complex auditory comprehension into three incremental stages: Warm-up (MCQ/QA) $\rightarrow$ Capability Ramp-up $\rightarrow$ Final Alignment (Full JSON).
|
| 13 |
|