|
|
--- |
|
|
license: apache-2.0 |
|
|
datasets: |
|
|
- stapesai/ssi-speech-emotion-recognition |
|
|
language: |
|
|
- en |
|
|
base_model: |
|
|
- facebook/wav2vec2-base-960h |
|
|
pipeline_tag: audio-classification |
|
|
library_name: transformers |
|
|
tags: |
|
|
- emotion |
|
|
- audio |
|
|
- classification |
|
|
- music |
|
|
- facebook |
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
# Speech-Emotion-Classification |
|
|
|
|
|
> **Speech-Emotion-Classification** is a fine-tuned version of `facebook/wav2vec2-base-960h` for **multi-class audio classification**, specifically trained to detect **emotions** in speech. This model utilizes the `Wav2Vec2ForSequenceClassification` architecture to accurately classify speaker emotions from audio signals. |
|
|
|
|
|
> \[!note] |
|
|
> Wav2Vec2: Self-Supervised Learning for Speech Recognition |
|
|
> [https://arxiv.org/pdf/2006.11477](https://arxiv.org/pdf/2006.11477) |
|
|
|
|
|
|
|
|
```py |
|
|
Classification Report: |
|
|
|
|
|
precision recall f1-score test_support |
|
|
|
|
|
Anger 0.8314 0.9346 0.8800 306 |
|
|
Calm 0.7949 0.8857 0.8378 35 |
|
|
Disgust 0.8261 0.8287 0.8274 321 |
|
|
Fear 0.8303 0.7377 0.7812 305 |
|
|
Happy 0.8929 0.7764 0.8306 322 |
|
|
Neutral 0.8423 0.9303 0.8841 287 |
|
|
Sad 0.7749 0.7825 0.7787 308 |
|
|
Surprised 0.9478 0.9478 0.9478 115 |
|
|
|
|
|
accuracy 0.8379 1999 |
|
|
macro avg 0.8426 0.8530 0.8460 1999 |
|
|
weighted avg 0.8392 0.8379 0.8367 1999 |
|
|
``` |
|
|
|
|
|
 |
|
|
|
|
|
 |
|
|
|
|
|
--- |
|
|
|
|
|
## Label Space: 8 Classes |
|
|
|
|
|
``` |
|
|
Class 0: Anger |
|
|
Class 1: Calm |
|
|
Class 2: Disgust |
|
|
Class 3: Fear |
|
|
Class 4: Happy |
|
|
Class 5: Neutral |
|
|
Class 6: Sad |
|
|
Class 7: Surprised |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## Install Dependencies |
|
|
|
|
|
```bash |
|
|
pip install gradio transformers torch librosa hf_xet |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## Inference Code |
|
|
|
|
|
```python |
|
|
import gradio as gr |
|
|
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor |
|
|
import torch |
|
|
import librosa |
|
|
|
|
|
# Load model and processor |
|
|
model_name = "prithivMLmods/Speech-Emotion-Classification" |
|
|
model = Wav2Vec2ForSequenceClassification.from_pretrained(model_name) |
|
|
processor = Wav2Vec2FeatureExtractor.from_pretrained(model_name) |
|
|
|
|
|
# Label mapping |
|
|
id2label = { |
|
|
"0": "Anger", |
|
|
"1": "Calm", |
|
|
"2": "Disgust", |
|
|
"3": "Fear", |
|
|
"4": "Happy", |
|
|
"5": "Neutral", |
|
|
"6": "Sad", |
|
|
"7": "Surprised" |
|
|
} |
|
|
|
|
|
def classify_audio(audio_path): |
|
|
# Load and resample audio to 16kHz |
|
|
speech, sample_rate = librosa.load(audio_path, sr=16000) |
|
|
|
|
|
# Process audio |
|
|
inputs = processor( |
|
|
speech, |
|
|
sampling_rate=sample_rate, |
|
|
return_tensors="pt", |
|
|
padding=True |
|
|
) |
|
|
|
|
|
with torch.no_grad(): |
|
|
outputs = model(**inputs) |
|
|
logits = outputs.logits |
|
|
probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist() |
|
|
|
|
|
prediction = { |
|
|
id2label[str(i)]: round(probs[i], 3) for i in range(len(probs)) |
|
|
} |
|
|
|
|
|
return prediction |
|
|
|
|
|
# Gradio Interface |
|
|
iface = gr.Interface( |
|
|
fn=classify_audio, |
|
|
inputs=gr.Audio(type="filepath", label="Upload Audio (WAV, MP3, etc.)"), |
|
|
outputs=gr.Label(num_top_classes=8, label="Emotion Classification"), |
|
|
title="Speech Emotion Classification", |
|
|
description="Upload an audio clip to classify the speaker's emotion from voice signals." |
|
|
) |
|
|
|
|
|
if __name__ == "__main__": |
|
|
iface.launch() |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
|
|
|
|
|
|
## Original Label |
|
|
|
|
|
```py |
|
|
"id2label": { |
|
|
"0": "ANG", |
|
|
"1": "CAL", |
|
|
"2": "DIS", |
|
|
"3": "FEA", |
|
|
"4": "HAP", |
|
|
"5": "NEU", |
|
|
"6": "SAD", |
|
|
"7": "SUR" |
|
|
}, |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
`Speech-Emotion-Classification` is designed for: |
|
|
|
|
|
* **Speech Emotion Analytics** – Analyze speaker emotions in call centers, interviews, or therapeutic sessions. |
|
|
* **Conversational AI Personalization** – Adjust voice assistant responses based on detected emotion. |
|
|
* **Mental Health Monitoring** – Support emotion recognition in voice-based wellness or teletherapy apps. |
|
|
* **Voice Dataset Curation** – Tag or filter speech datasets by emotion for research or model training. |
|
|
* **Media Annotation** – Automatically annotate podcasts, audiobooks, or videos with speaker emotion metadata. |
