Commit ·
24d876f
1
Parent(s): 9b5a816
Update README
Browse files
README.md
CHANGED
|
@@ -1,4 +1,53 @@
|
|
| 1 |
---
|
| 2 |
language:
|
| 3 |
- en
|
| 4 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
language:
|
| 3 |
- en
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
## Configurations:
|
| 7 |
+
|
| 8 |
+
- Use ESPNet's default frontend to extract features. The sampling rate is 8000 Hz, with a frame length of 25 ms and a frame shift of 10 ms. The frontend extracts 23 log-scaled Mel-filterbanks.
|
| 9 |
+
- Follow the frame concatenation and subsampling strategy described in paper [[2]]. Each frame is concatenated with the preceding and following 7 frames, followed by subsampling with a factor of 10. As a result, a 345-dimensional acoustic feature (23 × 15) is extracted for each 100 ms.
|
| 10 |
+
- Training and testing are performed exclusively on data with 4 speakers.
|
| 11 |
+
- Use 4 layer stacked Transformer encoder, each outputs 256-dimensional frame-wise embeddings.
|
| 12 |
+
- The training process spans 500 epochs.
|
| 13 |
+
- Detailed configurations are defined in `exp/diar/train_diar_diar_raw/config.yaml`.
|
| 14 |
+
|
| 15 |
+
## RESULTS
|
| 16 |
+
### Environments
|
| 17 |
+
- date: `Thu Dec 19 22:03:53 EST 2024`
|
| 18 |
+
- python version: `3.11.10 (main, Oct 3 2024, 07:29:13) [GCC 11.2.0]`
|
| 19 |
+
- espnet version: `espnet 202409`
|
| 20 |
+
- pytorch version: `pytorch 2.4.0`
|
| 21 |
+
- Git hash: `c12b3d59ca4fd8847edf274e56a1716474d2a30e`
|
| 22 |
+
- Commit date: `Thu Dec 19 21:58:26 2024 -0500`
|
| 23 |
+
|
| 24 |
+
### diar_train_diar_raw
|
| 25 |
+
#### DER
|
| 26 |
+
diarized_test
|
| 27 |
+
|threshold_median_collar|DER|
|
| 28 |
+
|---|---|
|
| 29 |
+
|result_th0.3_med11_collar0.0|71.73|
|
| 30 |
+
|result_th0.3_med1_collar0.0|74.62|
|
| 31 |
+
|result_th0.4_med11_collar0.0|70.10|
|
| 32 |
+
|result_th0.4_med1_collar0.0|71.98|
|
| 33 |
+
|result_th0.5_med11_collar0.0|70.57|
|
| 34 |
+
|result_th0.5_med1_collar0.0|72.44|
|
| 35 |
+
|result_th0.6_med11_collar0.0|72.64|
|
| 36 |
+
|result_th0.6_med1_collar0.0|74.63|
|
| 37 |
+
|result_th0.7_med11_collar0.0|76.52|
|
| 38 |
+
|result_th0.7_med1_collar0.0|78.41|
|
| 39 |
+
### diar_train_diar_raw
|
| 40 |
+
#### DER
|
| 41 |
+
diarized_dev
|
| 42 |
+
|threshold_median_collar|DER|
|
| 43 |
+
|---|---|
|
| 44 |
+
|result_th0.3_med11_collar0.0|75.88|
|
| 45 |
+
|result_th0.3_med1_collar0.0|78.21|
|
| 46 |
+
|result_th0.4_med11_collar0.0|71.45|
|
| 47 |
+
|result_th0.4_med1_collar0.0|73.32|
|
| 48 |
+
|result_th0.5_med11_collar0.0|70.53|
|
| 49 |
+
|result_th0.5_med1_collar0.0|72.34|
|
| 50 |
+
|result_th0.6_med11_collar0.0|72.03|
|
| 51 |
+
|result_th0.6_med1_collar0.0|73.96|
|
| 52 |
+
|result_th0.7_med11_collar0.0|76.66|
|
| 53 |
+
|result_th0.7_med1_collar0.0|78.33|
|