File size: 1,159 Bytes
51653c7
 
455558e
 
9bfdfc8
 
 
 
 
 
 
455558e
01443a8
455558e
 
 
 
 
 
51653c7
53990c6
51653c7
 
 
53990c6
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
---
license: apache-2.0
---
# Qhash-v0.1
<div align="center">
<img src="https://huggingface.co/datasets/Quantamhash/Assets/resolve/main/images/dark_logo.png"
     alt="Title card" 
     style="width: 500px;
            height: auto;
            object-position: center top;">
</div>
<div align="center">
<img src="https://huggingface.co/datasets/Quantamhash/Assets/resolve/main/images/Qhash.png"
     alt="Title card" 
     style="width: 500px;
            height: auto;
            object-position: center top;">
</div>

---
This repository contains the speaker embedding models for our Qhash-v0.1 [transformer](Quantamhash/Qhash-v0.1-transformer) and [hybrid](Quantamhash/Qhash-v0.1-hybrid) models.

The speaker embedding models are based on the [ResNet293-SimAM-ASP](https://github.com/VoxBlink2/ScriptsForVoxBlink2/tree/main/asv) models from VoxBlink2. We use the pretrain models as we found the finetunes performed worse.

The output of the speaker embedding model is then passed through an LDA layer and compressed from 256 to 128 dimensions to remove further spurious information about the speaker embedding clip before being fed into the Qhash models.