whisper-small-diarization-0.2

anakib1/whisper-small-diarization-0.2

whisper-small-diarization-0.2 is a machine learning model from anakib1. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as apache-2.0.

should probably proofread and complete it, then remove this comment. --

Downloads · 30 days

17% of all-time downloads

All-time downloads

Public

Parameters

242M

8.7 GB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors967 MB · 100%

At a glance

Library: transformers
License: apache-2.0
Model type: whisper
Access: Public
Created: Jan 15, 2024
Updated: Jan 16, 2024
SHA: 8596eaf6

Base models

openai/whisper-small

Model card

From the Hugging Face model README

whisper-small-diarization-0.2

This model is a fine-tuned version of openai/whisper-small on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.4057
Speech Scored: 802.6678
Speech Miss: 209.2589
Speech Falarm: 9.0497
Speaker Miss: 412.5650
Speaker Falarm: 185.8256
Speaker Error: 153.1979
Speaker Correct: 1198.4045
Diarization Error: 751.5885
Frames: 1500.0
Speaker Wide Frames: 1564.7402
Speech Scored Ratio: 0.5351
Speech Miss Ratio: 0.1395
Speech Falarm Ratio: 0.0060
Speaker Correct Ratio: 0.7989
Speaker Miss Ratio: 0.2382
Speaker Falarm Ratio: 0.1203
Speaker Error Ratio: 0.0831
Diarization Error Ratio: 0.4416

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e-05
train_batch_size: 16
eval_batch_size: 24
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.05
num_epochs: 10
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Speech Scored	Speech Miss	Speech Falarm	Speaker Miss	Speaker Falarm	Speaker Error	Speaker Correct	Diarization Error	Frames	Speaker Wide Frames	Speech Scored Ratio	Speech Miss Ratio	Speech Falarm Ratio	Speaker Correct Ratio	Speaker Miss Ratio	Speaker Falarm Ratio	Speaker Error Ratio	Diarization Error Ratio
0.4753	1.0	225	0.4855	731.3740	280.5527	69.0541	614.8317	199.7986	151.0314	1127.7690	965.6617	1500.0	1564.7402	0.4876	0.1870	0.0460	0.7518	0.3772	0.1827	0.0796	0.6395
0.4572	2.0	450	0.4857	642.0541	369.8727	29.7105	773.0375	89.8413	131.1212	1124.9596	994.0	1500.0	1564.7402	0.4280	0.2466	0.0198	0.7500	0.4566	0.0839	0.0709	0.6114
0.4638	3.0	675	0.4618	748.0445	263.8823	43.8745	466.9677	356.8117	116.3025	1147.8718	940.0820	1500.0	1564.7402	0.4987	0.1759	0.0292	0.7652	0.2922	0.2364	0.0631	0.5917
0.4423	4.0	900	0.4477	740.9529	270.9738	30.4473	509.6905	235.4464	132.9747	1162.9712	878.1116	1500.0	1564.7402	0.4940	0.1806	0.0203	0.7753	0.3072	0.1581	0.0712	0.5365
0.4164	5.0	1125	0.4309	737.6809	274.2459	12.4037	512.4150	173.0157	138.8300	1178.9698	824.2607	1500.0	1564.7402	0.4918	0.1828	0.0083	0.7860	0.3010	0.1130	0.0754	0.4893
0.3924	6.0	1350	0.4112	812.0453	199.8814	13.5414	382.1125	253.1543	140.2999	1194.7111	775.5667	1500.0	1564.7402	0.5414	0.1333	0.0090	0.7965	0.2235	0.1713	0.0755	0.4702
0.3765	7.0	1575	0.4085	806.7515	205.1752	12.1369	405.6992	202.0323	149.4699	1197.7762	757.2014	1500.0	1564.7402	0.5378	0.1368	0.0081	0.7985	0.2361	0.1250	0.0829	0.4439
0.3814	8.0	1800	0.4051	802.6016	209.3252	9.5911	398.2677	213.9582	144.1378	1199.8329	756.3636	1500.0	1564.7402	0.5351	0.1396	0.0064	0.7999	0.2367	0.1275	0.0794	0.4436
0.3965	9.0	2025	0.4111	768.8736	243.0532	6.9250	474.9355	148.3069	146.9695	1194.2729	770.2119	1500.0	1564.7402	0.5126	0.1620	0.0046	0.7962	0.2742	0.0932	0.0806	0.4480
0.4048	10.0	2250	0.4057	802.6678	209.2589	9.0497	412.5650	185.8256	153.1979	1198.4045	751.5885	1500.0	1564.7402	0.5351	0.1395	0.0060	0.7989	0.2382	0.1203	0.0831	0.4416

Framework versions

Transformers 4.36.2
Pytorch 2.0.0
Datasets 2.16.1
Tokenizers 0.15.0

Library: transformers
Type: whisper
License: apache-2.0
Created: Jan 15, 2024
Updated: Jan 16, 2024