wav2vec2-commonvoice-ja-finetuned

Fine-tuned facebook/wav2vec2-large-xlsr-53 for Japanese ASR using Common Voice Scripted Speech 24.0 - Japanese.

• Task: Automatic Speech Recognition (CTC)
• Language: Japanese
• License: apache-2.0
• Base model: facebook/wav2vec2-large-xlsr-53
• Dataset: Common Voice Scripted Speech 24.0 - Japanese
• Validation CER (best): 0.3107
• Decoding: greedy CTC, without external language model

Usage

import torch
import librosa
from transformers import AutoModelForCTC, Wav2Vec2CTCTokenizer, Wav2Vec2FeatureExtractor, Wav2Vec2Processor

repo_id = "takehika/wav2vec2-commonvoice-ja-finetuned"

tokenizer = Wav2Vec2CTCTokenizer.from_pretrained(repo_id)
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(repo_id)
processor = Wav2Vec2Processor(feature_extractor=feature_extractor, tokenizer=tokenizer)
model = AutoModelForCTC.from_pretrained(repo_id)

speech, _ = librosa.load("sample.wav", sr=16_000)
inputs = processor(speech, sampling_rate=16_000, return_tensors="pt", padding=True)

with torch.no_grad():
    logits = model(**inputs).logits

pred_ids = torch.argmax(logits, dim=-1)
text = processor.batch_decode(pred_ids)[0]
print(text)

Data

• Dataset: Common Voice Scripted Speech 24.0 - Japanese

Notes: Text normalization was applied (NFKC normalization, selected symbol removal, whitespace cleanup).

Training

• Batch size: 4
• Gradient accumulation: 2 (effective 8)
• Learning rate: 2e-5
• Scheduler: constant_with_warmup
• Warmup steps: 500
• Eval/save steps: 500

Evaluation

• Validation CER : 0.3107

Intended Use & Limitations

• Intended use: Japanese ASR for 16 kHz speech.
• Domain shift: performance may degrade for noisy audio, strong dialect/accent variation, telephone speech, or overlapping speakers.
• Text style: punctuation handling and formatting may differ from natural writing due to CTC-style decoding and preprocessing.
• Reliability: outputs may include transcription errors; human review is recommended for high-stakes use.

Attribution & Licenses

• License: Apache-2.0
• Base model facebook/wav2vec2-large-xlsr-53: Apache-2.0
• Dataset (Common Voice Scripted Speech 24.0 - Japanese): CC0-1.0

References:

• Base model card: https://huggingface.co/facebook/wav2vec2-large-xlsr-53
• Common Voice datasets portal: https://datacollective.mozillafoundation.org/datasets