GitTaskBench / test_scripts /FunASR_02 /test_script.py
Nicole-Yi's picture
Upload 103 files
4a5e815 verified
Raw
History Blame Contribute Delete
4.44 kB
import os
import sys
import json
import argparse
from difflib import SequenceMatcher
import datetime
import re
def check_file_exists(file_path):
"""Check if file exists and is not empty"""
if not os.path.exists(file_path):
return False, f"File does not exist: {file_path}"
if os.path.getsize(file_path) == 0:
return False, f"File is empty: {file_path}"
return True, ""
def cer(ref, hyp):
"""Character Error Rate = Edit Distance / Length of Reference"""
import numpy as np
ref = list(ref)
hyp = list(hyp)
d = np.zeros((len(ref)+1, len(hyp)+1), dtype=int)
for i in range(len(ref)+1):
d[i][0] = i
for j in range(len(hyp)+1):
d[0][j] = j
for i in range(1, len(ref)+1):
for j in range(1, len(hyp)+1):
cost = 0 if ref[i-1] == hyp[j-1] else 1
d[i][j] = min(
d[i-1][j] + 1, # deletion
d[i][j-1] + 1, # insertion
d[i-1][j-1] + cost # substitution
)
return d[len(ref)][len(hyp)] / max(len(ref), 1)
def is_likely_english(text):
english_letters = re.findall(r'[a-zA-Z]', text)
if not english_letters:
return False
ratio = len(english_letters) / max(len(text), 1)
return ratio > 0.5 and len(english_letters) >= 10 # at least 10 letters, >50% are English
def load_transcripts(file_path):
"""Load transcript text from file"""
try:
with open(file_path, 'r', encoding='utf-8') as f:
return f.read().replace("\n", ""), ""
except Exception as e:
return None, str(e)
def evaluate(system_output_file, ground_truth_file, cer_threshold=0.05):
"""Main evaluation function: Calculate CER between system output and ground truth"""
# Check files
process_ok, process_msg = check_file_exists(system_output_file)
if not process_ok:
return False, False, process_msg
process_ok, process_msg = check_file_exists(ground_truth_file)
if not process_ok:
return False, False, process_msg
# Load transcripts
system_trans, msg = load_transcripts(system_output_file)
if system_trans is None:
return True, False, f"Failed to load system output: {msg}"
ground_truth, msg = load_transcripts(ground_truth_file)
if ground_truth is None:
return True, False, f"Failed to load ground truth: {msg}"
if not is_likely_english(system_trans):
return True, False, "Output text does not appear to be valid English transcription"
# Calculate CER
score = cer(ground_truth, system_trans)
comments = [f"CER = {score:.4f}"]
result_ok = score <= cer_threshold
if not result_ok:
comments.append(f"CER ({score:.4f}) exceeds threshold {cer_threshold}")
return True, result_ok, "\n".join(comments)
def save_results_to_jsonl(process_ok, result_ok, comments, jsonl_file):
"""Save test results to JSONL file"""
current_time = datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
result_data = {
"Process": bool(process_ok),
"Result": bool(result_ok),
"TimePoint": current_time,
"comments": comments
}
os.makedirs(os.path.dirname(jsonl_file), exist_ok=True)
with open(jsonl_file, 'a', encoding='utf-8') as f:
json.dump(result_data, f, ensure_ascii=False, default=str)
f.write('\n')
def main():
parser = argparse.ArgumentParser(description='Evaluate speech recognition results')
parser.add_argument('--output', required=True, help='System output file path')
parser.add_argument('--groundtruth', required=True, help='Ground truth file path')
parser.add_argument('--cer_threshold', type=float, default=0.10, help='CER threshold')
parser.add_argument('--result', required=True, help='Result JSONL file path')
args = parser.parse_args()
process_ok, result_ok, comments = evaluate(
args.output,
args.groundtruth,
args.cer_threshold
)
save_results_to_jsonl(process_ok, result_ok, comments, args.result)
if not process_ok:
print(f"Processing failed: {comments}")
if not result_ok:
print(f"Results do not meet requirements: {comments}")
print("Test completed") # Changed to neutral prompt
if __name__ == "__main__":
main()