Malaya-speech_fine-tune_MrBrown_15_Jun Open Source Model - Free Realization of Singapore English Speech Recognition

Malaya Speech Fine Tune MrBrown 15 Jun

Developed by RuiqianLi

This model is a fine-tuned version of malay-huggingface/wav2vec2-xls-r-300m-mixed on the uob_singlish dataset, primarily used for Singapore English speech recognition.

Speech Recognition

Transformers

#Singapore English Speech Recognition #Low-resource Speech Optimization #Small-batch Fine-tuning

Downloads 44

Release Time : 6/15/2022

Model Overview

This is a speech recognition model optimized for Singapore English (Singlish), fine-tuned based on the wav2vec2-xls-r-300m-mixed architecture.

Model Features

Singapore English Optimization

Specifically optimized for the dialectal features of Singapore English (Singlish)

Efficient Fine-tuning

Achieves good results on limited data through efficient fine-tuning based on a pre-trained model

Low Word Error Rate

Achieves a word error rate (WER) of 24.49% on the evaluation set

Model Capabilities

Singapore English Speech Recognition

Continuous Speech-to-Text

Dialect Speech Processing

Use Cases

Speech Transcription

Singapore English Dialogue Transcription

Transcribes dialogue content featuring Singapore English characteristics into text

Word error rate 24.49%

Voice Assistants

Dialect Voice Assistant

Supports voice interaction systems for Singapore English

🚀 Malaya-speech_fine-tune_MrBrown_15_Jun

This model is a fine - tuned version of [malay - huggingface/wav2vec2 - xls - r - 300m - mixed](https://huggingface.co/malay - huggingface/wav2vec2 - xls - r - 300m - mixed) on the uob_singlish dataset. It offers significant improvements in speech - related tasks on this specific dataset.

🚀 Quick Start

This model is a fine - tuned version of [malay - huggingface/wav2vec2 - xls - r - 300m - mixed](https://huggingface.co/malay - huggingface/wav2vec2 - xls - r - 300m - mixed) on the uob_singlish dataset. It achieves the following results on the evaluation set:

Loss: 0.4822
Wer: 0.2449

📚 Documentation

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

🔧 Technical Details

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0001
train_batch_size: 2
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 4
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e - 08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 1000
num_epochs: 100
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1607	5.26	200	0.3983	0.2381
0.5184	10.52	400	0.3256	0.2245
0.2993	15.78	600	0.3437	0.2426
0.2485	21.05	800	0.4547	0.2585
0.1917	26.31	1000	0.4598	0.2517
0.1586	31.57	1200	0.4050	0.2290
0.1486	36.83	1400	0.4186	0.2653
0.1307	42.1	1600	0.4284	0.2857
0.0895	47.36	1800	0.5158	0.2562
0.0526	52.62	2000	0.4525	0.2449
0.0553	57.88	2200	0.4364	0.2336
0.037	63.16	2400	0.3873	0.2449
0.0439	68.42	2600	0.3914	0.2404
0.0411	73.68	2800	0.4673	0.2494
0.0242	78.94	3000	0.4801	0.2426
0.0833	84.21	3200	0.4641	0.2630
0.034	89.47	3400	0.4607	0.2449
0.02	94.73	3600	0.4825	0.2449
0.0211	99.99	3800	0.4822	0.2449

Framework versions

Transformers 4.11.3
Pytorch 1.10.0+cu113
Datasets 1.18.3
Tokenizers 0.10.3

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