filename_prep = re.search(r"(?<=system-output\/)(.*?)(?=\.txt)", infile).group(0)
outfile = "./results/google/universal/google-" + filename_prep + ".csv"
trans_file = "./results/google/system-trans-text/google-" + filename_prep + "-trans.txt"
# setting initial utterance as jiwer can't handle empty strings.
# tsoft = the start of the file.
prev = "tsotf"
utt = ""
# Google specific processing.
# This function extracts each new hypothesis with its time and processes it.
# Simultaneously, finalised hypotheses are stored for final WER calculations.
with open(infile, 'r') as f:
for line in f:
if line.startswith("Finished"):
fin = re.search(r"(?<=Finished: )(.*)(?=\n)", line).group(0)
if line.startswith("Time"):
time = re.search(r"(?<=Time: )(.*)(?=\n)", line).group(0)
if line.startswith("Transcript"):
utt = re.search(r"(?<=Transcript: )(.*)(?=\n)", line).group(0)
utt = utt.replace(".", "")
if fin == "False":
process(outfile, time, prev, utt)
prev = utt
else:
process(outfile, time, prev, utt)
add_trans_chunk(trans_file, utt.lower())
prev = "tsotf"
# Universal output finalised.
clean_csv(outfile)
from universal import add_trans_chunk
import sys
import re
# Given a system output and which channel it is from, the final transcript is stored.
# This is to calculate the WER.
infile = sys.argv[1]
side = sys.argv[2]
if side == "left":
filename_prep = re.search(r"(?<=left\/)(.*?)(?=\.txt)", infile).group(0)
elif side == "right":
filename_prep = re.search(r"(?<=right\/)(.*?)(?=\.txt)", infile).group(0)
else:
print("Which side?")
trans_file = "./results/msoft/split-system-trans-text/" + side + "-" + filename_prep + "-trans.txt"
with open(infile, 'r') as f:
for line in f:
if line.startswith("JSON"):
transcript = re.search(r"(?<=DisplayText\":\")(.*?)(?=\")", line)
if transcript:
transcript = transcript.group(0)
add_trans_chunk(trans_file, transcript.lower())
# IBM specific processing.
# This function extracts each new hypothesis with its time and processes it.
# Simultaneously, finalised hypotheses are stored for final WER calculations.
with open(infile, 'r') as f:
for line in f:
check = line.replace(" ", "").replace("%HESITATION", "")
try:
number = re.search(r"[+-]?([0-9]*[.])?[0-9]+", line).group(0)
except AttributeError:
pass
if check.startswith("\"transcript"):
utt = re.search(r"(?<=transcript\"\: \")(.*?)(?=\")",
line).group(0)
utt = utt.lower().replace("%hesitation", "")
if check.startswith(number) and not "," in line:
time = check.replace("\n", "")
if check.startswith("\"final"):
if not utt.isspace():
# print(utt)
if "false" in line:
process(outfile, time, prev, utt)
prev = utt
else:
process(outfile, time, prev, utt)
add_trans_chunk(trans_file, utt)
prev = "tsotf"
# Universal output finalised.
clean_csv(outfile)