Soxan

در زبان پارسی به نام سخن

This repository consists of models, scripts, and notebooks that help you to use all the benefits of Wav2Vec 2.0 in your research. In the following, I'll show you how to train speech tasks in your dataset and how to use the pretrained models.

How to train

I'm just at the beginning of all the possible speech tasks. To start, we continue the training script with the speech emotion recognition problem.

Training - Notebook

Task	Notebook
Speech Emotion Recognition (Wav2Vec 2.0)
Speech Emotion Recognition (Hubert)
Audio Classification (Wav2Vec 2.0)

Training - CMD

python3 run_wav2vec_clf.py \
    --pooling_mode="mean" \
    --model_name_or_path="lighteternal/wav2vec2-large-xlsr-53-greek" \
    --model_mode="wav2vec2" \ # or you can use hubert
    --output_dir=/path/to/output \
    --cache_dir=/path/to/cache/ \
    --train_file=/path/to/train.csv \
    --validation_file=/path/to/dev.csv \
    --test_file=/path/to/test.csv \
    --per_device_train_batch_size=4 \
    --per_device_eval_batch_size=4 \
    --gradient_accumulation_steps=2 \
    --learning_rate=1e-4 \
    --num_train_epochs=5.0 \
    --evaluation_strategy="steps"\
    --save_steps=100 \
    --eval_steps=100 \
    --logging_steps=100 \
    --save_total_limit=2 \
    --do_eval \
    --do_train \
    --fp16 \
    --freeze_feature_extractor

Prediction

import torch
import torch.nn as nn
import torch.nn.functional as F
import torchaudio
from transformers import AutoConfig, Wav2Vec2FeatureExtractor
from src.models import Wav2Vec2ForSpeechClassification, HubertForSpeechClassification

model_name_or_path = "path/to/your-pretrained-model"

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
config = AutoConfig.from_pretrained(model_name_or_path)
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_name_or_path)
sampling_rate = feature_extractor.sampling_rate

# for wav2vec
model = Wav2Vec2ForSpeechClassification.from_pretrained(model_name_or_path).to(device)

# for hubert
model = HubertForSpeechClassification.from_pretrained(model_name_or_path).to(device)

def speech_file_to_array_fn(path, sampling_rate):
    speech_array, _sampling_rate = torchaudio.load(path)
    resampler = torchaudio.transforms.Resample(_sampling_rate, sampling_rate)
    speech = resampler(speech_array).squeeze().numpy()
    return speech

def predict(path, sampling_rate):
    speech = speech_file_to_array_fn(path, sampling_rate)
    inputs = feature_extractor(speech, sampling_rate=sampling_rate, return_tensors="pt", padding=True)
    inputs = {key: inputs[key].to(device) for key in inputs}

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

    scores = F.softmax(logits, dim=1).detach().cpu().numpy()[0]
    outputs = [{"Emotion": config.id2label[i], "Score": f"{round(score * 100, 3):.1f}%"} for i, score in
               enumerate(scores)]
    return outputs

path = "/path/to/disgust.wav"
outputs = predict(path, sampling_rate)    

Output:

[
    {'Emotion': 'anger', 'Score': '0.0%'},
    {'Emotion': 'disgust', 'Score': '99.2%'},
    {'Emotion': 'fear', 'Score': '0.1%'},
    {'Emotion': 'happiness', 'Score': '0.3%'},
    {'Emotion': 'sadness', 'Score': '0.5%'}
]

Demos

Models