Simple Speech Recognition System

Dataset

• Dataset Name: speech-commands
• Number of Recognizable Words: 35
• Word Categories:
  1. Numeric: zero, one, two, three, four, five, six, seven, eight, nine
  2. Directional: left, right, forward, backward, up, down
  3. Command: go, stop, yes, no, on, off, follow
  4. Animal: bird, cat, dog
  5. Other: bed, house, happy, tree, wow, learn, visual, sheila, marvin

Model Files

• Model File: speech_commands_model_epoch_20_9621--64mel.pth
• Test Set Accuracy: 96.05%

Training Code

• Training Code File: train.py
• Function: Train a speech recognition model using the speech-commands dataset

Inference Code

• Inference Code File: Inference.ipynb
• Functions:
  1. Recognize the word corresponding to a single .wav audio file
  2. Recognize the words corresponding to all .wav files in a folder
  3. Record for 2 seconds and recognize the spoken word
  4. Continuously record and recognize a series of spoken words, providing the (start time, end time) for each word