Wav2vec2 Base 100h

🚀 Wav2Vec2-Base-100h

This is a fork of facebook/wav2vec2-base-100h, which focuses on audio and automatic speech recognition tasks using the LibriSpeech ASR dataset.

✨ Features

• Document reproducible evaluation to new transformer and datasets version.
• Use a batch size of 1 to reproduce results.
• Validated with transformers v4.15.0 and datasets 1.18.0.
• Manual installation of pypkg librosa and jiwer might be required.

📦 Installation

You may need to manually install the following pypkgs:

pip install librosa jiwer

💻 Usage Examples

Basic Usage

The following code snippet shows how to evaluate facebook/wav2vec2-base-100h on LibriSpeech's "clean" and "other" test data.

from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import soundfile as sf
import torch
from jiwer import wer

librispeech_eval = load_dataset("librispeech_asr", "clean", split="test")
# librispeech_eval = load_dataset("librispeech_asr", "other", split="test")

model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-100h").to("cuda")
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-100h")

def map_to_array(batch):
    # speech, _ = sf.read(batch["file"])
    # batch["speech"] = speech
    batch["speech"] = batch['audio']['array']
    return batch

librispeech_eval = librispeech_eval.map(map_to_array)

def map_to_pred(batch):
    input_values = processor(batch["speech"], return_tensors="pt", padding="longest").input_values
    with torch.no_grad():
        logits = model(input_values.to("cuda")).logits

    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = processor.batch_decode(predicted_ids)
    batch["transcription"] = transcription
    return batch

result = librispeech_eval.map(map_to_pred, batched=True, batch_size=1, remove_columns=["speech"])

print("WER:", wer(result["text"], result["transcription"]))

Results

📄 License

This project is licensed under the Apache 2.0 license.