XLS-R-300M-fr-0 Open-Source Automatic Speech Recognition Model - Free and Accurate French Speech to Text Conversion

Xls R 300m Fr 0

Developed by AlexN

This is an automatic speech recognition (ASR) model fine-tuned on the Common Voice 8.0 French dataset based on facebook/wav2vec2-xls-r-300m, supporting French speech-to-text tasks.

Speech Recognition

Transformers

FrenchOpen Source License:Apache-2.0 #French Speech Recognition #Multi-scenario Robustness #Low WER Optimization

Downloads 24

Release Time : 3/2/2022

Model Overview

This model is specifically designed for French speech recognition tasks and performs well on the Common Voice 8.0 French dataset with a WER of 36.81%.

Model Features

French Speech Recognition

Speech recognition capabilities optimized specifically for French.

Based on Large-scale Pretrained Model

Fine-tuned on the facebook/wav2vec2-xls-r-300m model, featuring powerful speech feature extraction capabilities.

Multi-dataset Validation

Validated on multiple datasets including Common Voice and Robust Speech Events.

Model Capabilities

French Speech-to-Text

Automatic Speech Recognition

Handling French speech with different accents

Use Cases

Speech Transcription

French Speech Transcription

Convert French speech content into text

WER 36.81% (Common Voice 8.0 test set)

Voice Assistants

French Voice Command Recognition

Used for voice command recognition in French voice assistant systems

🚀 XLS-R-300M - FR

This model is a fine - tuned speech recognition model. It is based on the pre - trained model facebook/wav2vec2-xls-r-300m and fine - tuned on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - FR dataset. It can achieve good performance in French speech recognition tasks.

🚀 Quick Start

This model is a fine - tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - FR dataset. It achieves the following results on the evaluation set:

Loss: 0.2388
Wer: 0.3681

📚 Documentation

Model Information

Property	Details
Model Type	Fine - tuned version of facebook/wav2vec2-xls-r-300m
Training Data	MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - FR

Evaluation Results

Task	Dataset	Metrics	Value
Speech Recognition	Common Voice 8.0 fr (mozilla - foundation/common_voice_8_0, args: fr)	Test WER	36.81
Automatic Speech Recognition	Robust Speech Event - Dev Data (speech - recognition - community - v2/dev_data, args: fr)	Test WER	35.55
Automatic Speech Recognition	Robust Speech Event - Test Data (speech - recognition - community - v2/eval_data, args: fr)	Test WER	39.94

Training Procedure

Training Hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0001
train_batch_size: 64
eval_batch_size: 64
seed: 42
optimizer: Adam with betas=(0.9, 0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 1500
num_epochs: 2.0
mixed_precision_training: Native AMP

Training Results

Training Loss	Epoch	Step	Validation Loss	Wer
4.3748	0.07	500	3.8784	1.0
2.8068	0.14	1000	2.8289	0.9826
1.6698	0.22	1500	0.8811	0.7127
1.3488	0.29	2000	0.5166	0.5369
1.2239	0.36	2500	0.4105	0.4741
1.1537	0.43	3000	0.3585	0.4448
1.1184	0.51	3500	0.3336	0.4292
1.0968	0.58	4000	0.3195	0.4180
1.0737	0.65	4500	0.3075	0.4141
1.0677	0.72	5000	0.3015	0.4089
1.0462	0.8	5500	0.2971	0.4077
1.0392	0.87	6000	0.2870	0.3997
1.0178	0.94	6500	0.2805	0.3963
0.992	1.01	7000	0.2748	0.3935
1.0197	1.09	7500	0.2691	0.3884
1.0056	1.16	8000	0.2682	0.3889
0.9826	1.23	8500	0.2647	0.3868
0.9815	1.3	9000	0.2603	0.3832
0.9717	1.37	9500	0.2561	0.3807
0.9605	1.45	10000	0.2523	0.3783
0.96	1.52	10500	0.2494	0.3788
0.9442	1.59	11000	0.2478	0.3760
0.9564	1.66	11500	0.2454	0.3733
0.9436	1.74	12000	0.2439	0.3747
0.938	1.81	12500	0.2411	0.3716
0.9353	1.88	13000	0.2397	0.3698
0.9271	1.95	13500	0.2388	0.3681

Framework Versions

Transformers 4.17.0.dev0
Pytorch 1.10.2+cu102
Datasets 1.18.2.dev0
Tokenizers 0.11.0

📄 License

This model is released 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