wav2vec2-base-common-voice-40p-persian-colab Open-source Model - Free Deployment to Realize Persian Speech-to-Text

Wav2vec2 Base Common Voice 40p Persian Colab

Developed by zoha

This model is a fine-tuned speech recognition model based on facebook/wav2vec2-base for Persian language datasets, primarily used for Persian speech-to-text tasks.

Speech Recognition

Transformers

Open Source License:Apache-2.0 #Persian speech recognition #Low-resource optimization #Word error rate optimization

Downloads 120

Release Time : 6/26/2022

Model Overview

This is an optimized Automatic Speech Recognition (ASR) model for Persian, based on the wav2vec2 architecture, fine-tuned for 40 epochs on the Common Voice Persian dataset.

Model Features

Persian Optimization

Specially optimized for the phonetic characteristics of Persian speech

Based on wav2vec2

Uses Facebook's open-source wav2vec2-base architecture as the base model

Medium-scale Training

Trained for 40 epochs, achieving a word error rate of 0.6024 on the validation set

Model Capabilities

Persian speech recognition

Speech-to-text

Use Cases

Speech Transcription

Persian Speech Transcription

Convert Persian speech content into text

Word error rate 0.6024

Voice Assistants

Persian Voice Command Recognition

Used for Persian voice assistant command recognition systems

🚀 wav2vec2-base-common-voice-40p-persian-colab

This model is a fine - tuned version of facebook/wav2vec2-base on the None dataset. It provides valuable results on the evaluation set, offering potential for speech - related tasks.

🚀 Quick Start

This model is a fine - tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:

Loss: 1.1805
Wer: 0.6024

📚 Documentation

Training and Evaluation

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.00018
train_batch_size: 16
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 2000
num_epochs: 40
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.9643	1.05	200	3.0107	1.0
2.7552	2.11	400	2.7370	0.9997
1.9144	3.16	600	1.8266	0.9703
1.502	4.21	800	1.3981	0.8996
1.3155	5.26	1000	1.2148	0.8507
0.9471	6.32	1200	1.1698	0.7860
0.8391	7.37	1400	1.1106	0.7857
0.7986	8.42	1600	1.1858	0.7769
0.7692	9.47	1800	1.1227	0.7603
0.7871	10.53	2000	1.0626	0.7612
0.6795	11.58	2200	1.1249	0.7209
0.4842	12.63	2400	1.1626	0.7336
0.492	13.68	2600	1.0995	0.7212
0.5117	14.74	2800	1.1406	0.7105
0.5649	15.79	3000	1.0603	0.6819
0.3232	16.84	3200	1.1781	0.7070
0.4098	17.89	3400	1.1182	0.6764
0.3917	18.95	3600	1.1320	0.6750
0.3712	20.0	3800	1.1920	0.6724
0.3157	21.05	4000	1.1102	0.6786
0.2397	22.11	4200	1.1924	0.6519
0.2751	23.16	4400	1.1497	0.6468
0.2279	24.21	4600	1.2274	0.6400
0.393	25.26	4800	1.1741	0.6436
0.1748	26.32	5000	1.2038	0.6327
0.1727	27.37	5200	1.1639	0.6347
0.255	28.42	5400	1.1948	0.6367
0.2261	29.47	5600	1.1560	0.6362
0.2359	30.53	5800	1.1227	0.6269
0.1668	31.58	6000	1.1861	0.6295
0.1699	32.63	6200	1.2442	0.6314
0.14	33.68	6400	1.1340	0.6277
0.1919	34.74	6600	1.1691	0.6139
0.2527	35.79	6800	1.1511	0.6110
0.1219	36.84	7000	1.2062	0.6139
0.1389	37.89	7200	1.2142	0.6072
0.135	38.95	7400	1.1967	0.6040
0.1563	40.0	7600	1.1805	0.6024

Framework versions

Transformers 4.11.3
Pytorch 1.10.0+cu113
Datasets 1.18.3
Tokenizers 0.10.3

📄 License

This model is licensed under the Apache - 2.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご