saansho/speech_emotion_recognition

Speech Emotion Recognition


A simple speech emotion recognition project that predicts the emotional state of a speaker from a WAV audio file.

The app uses:

• —A pre-trained TensorFlow/Keras model stored in models/final_industry_ser_model.keras
• —Audio preprocessing and MFCC feature extraction in src/
• —A Streamlit user interface in app/app.py

What this project does

1. 1.Loads an audio file in WAV format.
2. 2.Trims silence and normalizes the audio.
3. 3.Extracts MFCC features from the audio signal.
4. 4.Runs the features through a trained model.
5. 5.Shows the predicted emotion and confidence score in the web interface.

Project structure

• —app/ — Streamlit interface and file uploader
• —models/ — Pre-trained TensorFlow/Keras model file
• —src/ — Python modules for preprocessing, feature extraction, and prediction
• —dataset/ — optional audio dataset for testing or experimentation
• —outputs/ — generated graphs, reports, or other output files
• —requirements.txt — project dependencies
• —main.py — experimental pipeline script for batch processing audio files

Supported emotions

The model predicts one of these emotions:

• —angry
• —fearful
• —happy
• —neutral
• —sad

Requirements

This project is tested with:

• —Python 3.10+ or 3.11
• —tensorflow==2.16.2
• —streamlit==1.57.0
• —librosa==0.11.0
• —numpy==1.26.4
• —pandas==3.0.3
• —scikit-learn==1.8.0
• —soundfile==0.13.1

Note: For Hugging Face Spaces and Linux deployment, use tensorflow==2.16.2. Mac-specific packages like tensorflow-macos and tensorflow-metal are not required for Spaces.

Setup

1. 1.Create and activate a virtual environment:

python3.11 -m venv venv
source venv/bin/activate

1. 1.Install dependencies:

python -m pip install --upgrade pip setuptools wheel
pip install -r requirements.txt

Run the app

Start the Streamlit interface:

streamlit run app/app.py

Then open the URL shown by Streamlit in your browser.

Deploy to Hugging Face Spaces

This repository supports two deployment options.

Option 1: Native Streamlit Space

1. 1.Create a new Hugging Face Space.
2. 2.Choose Streamlit as the SDK.
3. 3.Push this repository to the Space.
4. 4.Ensure requirements.txt, app.py, and models/final_industry_ser_model.keras are included.

Hugging Face should detect the root app.py wrapper and run the Streamlit app automatically.

Option 2: Docker Space

1. 1.Create a new Hugging Face Space.
2. 2.Choose Docker as the SDK.
3. 3.Push this repository to the Space.

The provided Dockerfile installs ffmpeg and runs Streamlit on port 7860, so no further Docker configuration is required.

Deploy-ready commands

Use this sequence to push the repository and deploy it on Hugging Face Spaces:

git add .
git commit -m "Deploy Speech Emotion Recognition to Hugging Face Spaces"
git push origin main

Then create a new Hugging Face Space and choose one of these options:

• —Streamlit SDK: the root app.py wrapper will be detected automatically
• —Docker SDK: the existing Dockerfile installs ffmpeg and runs the app on port 7860

If you want to build locally first, make sure Docker Desktop is running and then use:

docker build -t ser-app .
docker run -p 7860:7860 ser-app

If the Docker daemon is not running, the build will fail with a socket connection error.

How to use the app

1. 1.Upload a .wav audio file using the file uploader.
2. 2.Click the Predict Emotion button.
3. 3.View the predicted emotion and confidence score.

Notes for developers

• —src/preprocessing.py handles audio loading, trimming, normalization, and length standardization.
• —src/feature_extraction.py extracts MFCC features from the audio.
• —src/prediction.py loads the model and returns the predicted emotion and confidence.
• —main.py is an experimental script for batch processing audio files from dataset/.

Future improvements

• —Add support for more audio file formats (MP3, FLAC)
• —Add real-time microphone recording
• —Expand emotion labels beyond the current five classes
• —Add model evaluation and result logging

License

This project is available under the MIT License.