Whisper-base-Arabic Open-Source Model - Focused on Arabic Speech-to-Text, Free and Accurate Recognition!

Whisper Base Arabic

Developed by YazanSalameh

An Arabic speech recognition model based on Whisper-base, fine-tuned on multiple Arabic datasets, specializing in Arabic speech-to-text tasks

Speech Recognition

Transformers

Supports Multiple LanguagesOpen Source License:Apache-2.0 #Arabic Speech-to-Text #Multi-dialect Support #Low-resource Optimization

Downloads 46

Release Time : 1/18/2024

Model Overview

This is an automatic speech recognition (ASR) model optimized for Arabic, based on OpenAI's Whisper-base architecture, supporting Arabic speech transcription into text

Model Features

Arabic Optimization

Specially optimized for Arabic speech characteristics, removing Arabic diacritics to improve recognition accuracy

Multi-dataset Training

Trained on a combination of Common Voice, MGB2, and Jordanian audio datasets, covering various Arabic dialects and accents

Efficient Training

Utilizes optimization techniques like gradient accumulation and linear learning rate scheduling for efficient training with limited resources

Model Capabilities

Arabic Speech Recognition

Speech-to-Text

Audio Transcription

Use Cases

Speech Transcription

Meeting Minutes

Automatically transcribe Arabic meeting recordings into text records

Word Error Rate 34.7%

Media Subtitle Generation

Automatically generate subtitles for Arabic video content

Voice Assistants

Arabic Voice Command Recognition

Used for Arabic voice assistants and smart home control

🚀 Whisper base arabic

This model is designed for Arabic speech - to - text transcription, achieving good performance on the evaluation set.

🚀 Quick Start

This model achieves the following results on the evaluation set:

Loss: 0.44
Wer: 34.7

✨ Features

Language Support: Supports Arabic language transcription.
Based on Whisper: Built upon the openai/whisper - base model.
Multiple Datasets: Trained on multiple datasets including mozilla - foundation/common_voice_16_0, BelalElhossany/mgb2_audios_transcriptions_non_overlap, and nadsoft/Jordan - Audio.

📦 Installation

No installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

No code examples are provided in the original document, so this section is skipped.

📚 Documentation

Training and evaluation data

Train set:

mozilla - foundation/common_voice_16_0 ar [train+validation]
BelalElhossany/mgb2_audios_transcriptions_non_overlap
nadsoft/Jordan - Audio

Cross - validation set: A total of 600 samples from the 3 sets were used to save time during training as the Colab free tier was used to train the model. Note: Evaluate accuracy in the way you see fit.

Training procedure

Arabic (حركات) was removed from the texts. The model was trained on the combined dataset for 6 epochs, and the best one was the fifth, so the model is basically the 5th epoch.

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e - 05
train_batch_size: 32
eval_batch_size: 16
seed: 42
gradient_accumulation_steps: 1
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4603	1	1437	0.4931	45.8857
0.2867	2	2874	0.4493	36.9973
0.2494	3	4311	0.4219	43.5553
0.1435	4	5748	0.4408	40.2351
0.1345	5	7185	0.4407	34.7081

🔧 Technical Details

Model Information

Property	Details
Model Type	Whisper base arabic
Base Model	openai/whisper - base
Training Datasets	mozilla - foundation/common_voice_16_0, BelalElhossany/mgb2_audios_transcriptions_non_overlap, nadsoft/Jordan - Audio
Evaluation Metrics	wer

Model Index

Name: Whisper base arabic
- Results:
  - Task:
    - Name: Automatic Speech Recognition
    - Type: automatic - speech - recognition
  - Metrics:
    - Name: Wer
    - Type: wer
    - Value: 34.7

📄 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