wav2vec2-large-xls-r-300m-Tatar

作者 kingabzpro · 下载 1 · 点赞 1 · 更新 2023-10-31

automatic-speech-recognition robust-speech-event hf-asr-leaderboard

wav2vec2-large-xls-r-300m-Tatar

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

Evaluation Commands

  1. To evaluate on mozilla-foundation/common_voice_8_0 with split test
python eval.py --model_id kingabzpro/wav2vec2-large-xls-r-300m-Tatar --dataset mozilla-foundation/common_voice_8_0 --config tt --split test

Inference With LM

import torch
from datasets import load_dataset
from transformers import AutoModelForCTC, AutoProcessor
import torchaudio.functional as F
model_id = "kingabzpro/wav2vec2-large-xls-r-300m-Tatar"
sample_iter = iter(load_dataset("mozilla-foundation/common_voice_8_0", "tt", split="test", streaming=True, use_auth_token=True))
sample = next(sample_iter)
resampled_audio = F.resample(torch.tensor(sample["audio"]["array"]), 48_000, 16_000).numpy()
model = AutoModelForCTC.from_pretrained(model_id)
processor = AutoProcessor.from_pretrained(model_id)
input_values = processor(resampled_audio, return_tensors="pt").input_values
with torch.no_grad():
    logits = model(input_values).logits
transcription = processor.batch_decode(logits.numpy()).text

Training hyperparameters

The following hyperparameters were used during training:

Training results

Training Loss Epoch Step Validation Loss Wer Cer
8.4116 12.19 500 3.4118 1.0 1.0
2.5829 24.39 1000 0.7150 0.6151 0.1582
0.4492 36.58 1500 0.5378 0.4577 0.1210
0.3007 48.77 2000 0.5068 0.4263 0.1117

Framework versions