Nicole-Yi's picture
Upload 103 files
4a5e815 verified
import argparse
import librosa
import numpy as np
from pesq import pesq
import os
import json
from datetime import datetime
def validate_wav_file(wav_path):
"""Validates if the WAV file exists, is non-empty, and is a valid audio file."""
try:
# Checking if file exists
if not os.path.exists(wav_path):
return False, f"File {wav_path} does not exist."
# Checking if file is non-empty
if os.path.getsize(wav_path) == 0:
return False, f"File {wav_path} is empty."
# Attempting to load the file to verify it's a valid WAV
librosa.load(wav_path, sr=16000)
return True, "Valid WAV file."
except Exception as e:
return False, f"Invalid WAV file {wav_path}: {str(e)}"
def evaluate_speech_enhancement(input_wav, output_wav, threshold=2.0):
# Loading audio files
ref, sr = librosa.load(input_wav, sr=16000) # Reference (noisy) audio
deg, sr = librosa.load(output_wav, sr=16000) # Enhanced (output) audio
# Ensuring same length for PESQ calculation
min_len = min(len(ref), len(deg))
ref = ref[:min_len]
deg = deg[:min_len]
# Calculating PESQ score (WB: Wideband, 16kHz)
pesq_score = pesq(16000, ref, deg, 'wb')
# Determining if task is successful based on threshold
task_passed = pesq_score >= threshold
return pesq_score, task_passed
def save_result_to_jsonl(process_status, result_status, comments, result_file):
"""Saves evaluation result to a JSONL file."""
result_entry = {
"Process": process_status,
"Result": result_status,
"TimePoint": datetime.now().strftime("%Y-%m-%dT%H:%M:%S"),
"comments": comments
}
print(comments)
# Writing to JSONL file in append mode with UTF-8 encoding
with open(result_file, 'a', encoding='utf-8') as f:
json.dump(result_entry, f, default=str)
f.write('\n')
def main():
# Setting up argument parser
parser = argparse.ArgumentParser(description='Evaluate speech enhancement using PESQ metric')
parser.add_argument('--groundtruth', type=str, required=True, help='Path to input (noisy) wav file')
parser.add_argument('--output', type=str, required=True, help='Path to output (enhanced) wav file')
parser.add_argument('--threshold', type=float, default=1.5, help='PESQ threshold for task success')
parser.add_argument('--result', type=str, required=True, help='Path to JSONL result file')
args = parser.parse_args()
# Validating input and output WAV files
input_valid, input_comment = validate_wav_file(args.groundtruth)
output_valid, output_comment = validate_wav_file(args.output)
process_status = input_valid and output_valid
comments = f"Input validation: {input_comment}; Output validation: {output_comment}"
pesq_score = None
task_passed = False
# Proceeding with evaluation only if both files are valid
if process_status:
try:
pesq_score, task_passed = evaluate_speech_enhancement(args.groundtruth, args.output, args.threshold)
comments += f"; PESQ Score: {pesq_score:.3f}, Task Passed: {task_passed} (Threshold: {args.threshold})"
except Exception as e:
comments += f"; Evaluation failed: {str(e)}"
task_passed = False
else:
task_passed = False
# Saving result to JSONL
save_result_to_jsonl(process_status, task_passed, comments, args.result)
# Printing results
print(f"PESQ Score: {pesq_score:.3f}" if pesq_score is not None else "Evaluation skipped due to invalid files")
print(f"Task Passed: {task_passed} (Threshold: {args.threshold})")
print(f"Results saved to: {args.result}")
if __name__ == "__main__":
main()