def create_voice_activity_clip(self, filename, outfile_name):
"""
1. Runs the voice activity detector on the clip
2. Calls the create_cutpoints_file program to create a file times listing where video should be cut
3. Combines the times listed on the cutpoints file to create a combined video file
"""
log.info("Running VAD on " + str(filename))
try:
v = VoiceActivityDetector(filename)
except Exception as e:
log.info("ERROR at VAD " + str(filename) + " " + e)
pass
# convert phase recognition array to second format
log.info("Converting windows to readible labels for " + str(filename))
try:
voice_activity_regions = v.convert_windows_to_readible_labels(v.detect_speech())
except:
log.debug("ERROR at convert windows to readable labels for file" + str(filename))
raise Exception
voice_activity_regions_array = []
# Flattening the list
for region in voice_activity_regions:
voice_activity_regions_array = voice_activity_regions_array + list(region.values())
if voice_activity_regions_array != []:
log.info("No voice activity detecvted for this clip.")
log.info("Finished creating array of voice activity regions")
try:
#generate a vector of seconds, which is where the array should be cut
cut_areas = self.get_most_dense_range(voice_activity_regions_array)
#creates a textfile.txt cut areas file in the wav directory
self.create_cutpoints_file(cut_areas, filename)
except Exception as e:
log.debug("could not create a text file with cut areas")
log.debug(e)
raise Exception
if outfile_name == None: #allows you to specify an outfile name, if none is given creates a name
file = "cut_stream" + filename[-11] + "_" + self.inputfile[:-4] + ".wav" # creates cut_name for files
else:
file = outfile_name
#creates a concatenated video file with high phrase density
try:
ffmpeg_calls.create_shortened_file(self.wav_directory, self.targetdirectory, file)
except subprocess.CalledProcessError as e:
log.debug(e)
log.info("Created: " + self.targetdirectory + "/" + file)
return file
def retrieve_confidence(id, filename):
#----- Retrieves Confidence For All Files Uploaded
log.info("Getting confidence for file: " + filename + " id: " + id)
language, confidence, response_json = VI_API.new_get_language(str(id))
log.info("File: " + filename + " language: " + str(language) + " confidence: " + str(confidence))
## -------- Reuploads video using a 5 min segment comprising random 20 second samples if the confidence is too low ----------------------
cnt = 2
try:
if confidence < 0.50:
log.info("Low confidence score, redoing processing -- this will take ~10 mins")
#Get the full audio filename corresponding to the clip
stream_number = str(filename.split("_")[1][-1])
encoded_file_name = stream_number + "output.wav"
encoded_file_path = wav_files_directory + "/" + encoded_file_name
file_length = ffmpeg_calls.retrieve_len(encoded_file_path)
#print("Failed to find video length")
new_language, new_confidence, new_json = 0, 0, {}
if file_length > MINIMUM_FILE_LENGTH_THRESHOLD:
# Take the shorter length
temp_cut_size = min(file_length, 2400)
log.info("Going to cut " + str(temp_cut_size) + " from the file")
#random_startpoint = random.randint(60, int(file_length - (temp_cut_size + 10)))
# generate 20 second ranges between the beginning of the video and the cutsize.
x = range(10,temp_cut_size, 20)
arr = random.sample(x, 15)
cutpoints_file = open(wav_files_directory + "/" + "cutpoints_random_sample.txt", "w")
num_cuts = int(MINIMUM_FILE_LENGTH_THRESHOLD / len(arr))
log.info("Number of cuts " + str(num_cuts))
for i in range(len(arr)):
cutpoints_file.write("file " + str(encoded_file_name) + "\n")
cutpoints_file.write("inpoint " + str(arr[i] - num_cuts) + "\n")
cutpoints_file.write("outpoint " + str(arr[i]) + "\n")
cutpoints_file.close()
log.info("Trying to create the shortened file from the cutpoints")
ffmpeg_calls.create_shortened_file(wav_files_directory, reprocessed_files_directory, filename, "cutpoints_random_sample.txt")
temp_id = VI_API.upload_video_file(filename, reprocessed_files_directory + "/" + filename)
new_language, new_confidence, new_json = VI_API.new_get_language(temp_id)
if new_confidence > confidence:
confidence = new_confidence
language = new_language
response_json = new_json
VI_API.clean_index([id])
else:
if temp_id != "None":
VI_API.clean_index([temp_id])
log.info("Writing the results to json ... ")
write_json_to_file(response_json, detection_results_json_directory + "/" + filename.split(".")[0] + ".json")
log.info("Wrote result to: " + detection_results_json_directory + "/" + filename.split(".")[0] + ".json")
return {
"streamName": filename,
"language": str(language),
"confidence": str(confidence),
"resultJsonFile": filename.split(".")[0] + ".json"
}
except Exception as e:
print("ERROR: Failed to get new confidence for file " + filename)
print e
return {}
def retrieve_confidence(id, filename):
#----- Retrieves Confidence For All Files Uploaded
log.info("Getting confidence for file: " + filename + " id: " + id)
language, confidence, response_json = VI_API.new_get_language(str(id))
log.info("File: " + filename + " language: " + str(language) + " confidence: " + str(confidence))
## -------- Reuploads video using a 5 min segment comprising random 20 second samples if the confidence is too low ----------------------
cnt = 2
try:
#If confidence is below a certain theshold
if confidence < 0.50:
log.info("Low confidence score, redoing processing -- this will take ~10 mins")
#Get the full audio filename corresponding to the clip
#This finds the streamnumber from the description of the file
#the wavfile format is 'cutstream2_output.wav', this takes the 2 from the filesname to find the stream number
stream_number = str(filename.split("_")[1][-1])
#This creates an encoded file name for the file, the encoded file name is simply the stream number + "output.wav". So stream 3 becomes 3output.wav
encoded_file_name = stream_number + "output.wav"
#This gives the full filepath the the newly created wav file
encoded_file_path = wav_files_directory + "/" + encoded_file_name
file_length = ffmpeg_calls.retrieve_len(encoded_file_path)
#print("Failed to find video length")
#initialize langauge and confidence at zero
new_language, new_confidence, new_json = 0, 0, {}
#If the file length is smaller than 5 minutes, don't do any processing
if file_length > MINIMUM_FILE_LENGTH_THRESHOLD:
# Take the shorter length
#The temp_cut_size of the file is the smaller between 20 minutes, and the length of the entire file
#If the file is smaller than 20 minutes, we will not take a random 20 minute sample from the video file and instead process the entire video file
temp_cut_size = min(file_length, 2400)
log.info("Going to cut " + str(temp_cut_size) + " from the file")
#random_startpoint = random.randint(60, int(file_length - (temp_cut_size + 10)))
# generate 20 second ranges between the beginning of the video and the cutsize.
x = range(10,temp_cut_size, 20)
#This takes a random sample from the 20 second segments created above, this is used to generate a random mix of 20 second segments to create a 5 minute file from
arr = random.sample(x, 15)
#Create a file with locations to cut to create a shorter file
cutpoints_file = open(wav_files_directory + "/" + "cutpoints_random_sample.txt", "w")
#Number of cuts is determined by file lenght / length of the array
num_cuts = int(MINIMUM_FILE_LENGTH_THRESHOLD / len(arr))
log.info("Number of cuts " + str(num_cuts))
#Creates a file called cutpoints .txt, which is a helper file used by ffmpeg to figure out where to cut the original file
for i in range(len(arr)):
cutpoints_file.write("file " + str(encoded_file_name) + "\n")
cutpoints_file.write("inpoint " + str(arr[i] - num_cuts) + "\n")
cutpoints_file.write("outpoint " + str(arr[i]) + "\n")
cutpoints_file.close()
log.info("Trying to create the shortened file from the cutpoints")
ffmpeg_calls.create_shortened_file(wav_files_directory, reprocessed_files_directory, filename, "cutpoints_random_sample.txt")
temp_id = VI_API.upload_video_file(filename, reprocessed_files_directory + "/" + filename)
new_language, new_confidence, new_json = VI_API.new_get_language(temp_id)
#If the new confidence generated by the new method is better than old confidence generated by the first method, use results from the new method
if new_confidence > confidence:
confidence = new_confidence
language = new_language
response_json = new_json
VI_API.clean_index([id])
else:
if temp_id != "None":
VI_API.clean_index([temp_id])
log.info("Writing the results to json ... ")
#This writes the results from audio detection to a JSON File
write_json_to_file(response_json, detection_results_json_directory + "/" + filename.split(".")[0] + ".json")
log.info("Wrote result to: " + detection_results_json_directory + "/" + filename.split(".")[0] + ".json")
#This returns the results in JSON Format
return {
"streamName": filename,
"language": str(language),
"confidence": str(confidence),
"resultJsonFile": filename.split(".")[0] + ".json"
}
except Exception as e:
print("ERROR: Failed to get new confidence for file " + filename)
print e
return {}