Update README.md
Browse files
README.md
CHANGED
|
@@ -1,4 +1,5 @@
|
|
| 1 |
---
|
|
|
|
| 2 |
license: other
|
| 3 |
language:
|
| 4 |
- en
|
|
@@ -9,6 +10,11 @@ tags:
|
|
| 9 |
- ASR
|
| 10 |
- chat
|
| 11 |
- voice
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
---
|
| 13 |
# Model Overview
|
| 14 |
|
|
@@ -68,10 +74,15 @@ Extensive evaluations confirm AF3’s effectiveness, setting new benchmarks on o
|
|
| 68 |
|
| 69 |
**This model is for non-commercial research purposes only.**
|
| 70 |
|
|
|
|
|
|
|
| 71 |
<center><img src="static/af3_radial-1.png" width="400"></center>
|
| 72 |
|
| 73 |
<br>
|
| 74 |
|
|
|
|
|
|
|
|
|
|
| 75 |
<center><img src="static/af3_main_diagram-1.png" width="800"></center>
|
| 76 |
|
| 77 |
|
|
|
|
| 1 |
---
|
| 2 |
+
arxiv: 2503.03983
|
| 3 |
license: other
|
| 4 |
language:
|
| 5 |
- en
|
|
|
|
| 10 |
- ASR
|
| 11 |
- chat
|
| 12 |
- voice
|
| 13 |
+
datasets:
|
| 14 |
+
- nvidia/LongAudio
|
| 15 |
+
- nvidia/AudioSkills
|
| 16 |
+
- nvidia/AF-Think
|
| 17 |
+
- nvidia/AF-Chat
|
| 18 |
---
|
| 19 |
# Model Overview
|
| 20 |
|
|
|
|
| 74 |
|
| 75 |
**This model is for non-commercial research purposes only.**
|
| 76 |
|
| 77 |
+
|
| 78 |
+
## Results:
|
| 79 |
<center><img src="static/af3_radial-1.png" width="400"></center>
|
| 80 |
|
| 81 |
<br>
|
| 82 |
|
| 83 |
+
## Model Architecture:
|
| 84 |
+
Audio Flamingo 3 uses AF-Whisper unified audio encoder, MLP-based audio adaptor, Decoder-only LLM backbone (Qwen2.5-7B), and Streaming TTS module (AF3-Chat). Audio Flamingo 3 can take up to 10 minutes of audio inputs.
|
| 85 |
+
|
| 86 |
<center><img src="static/af3_main_diagram-1.png" width="800"></center>
|
| 87 |
|
| 88 |
|