def test_deepspeech():
model_path = os.path.join("test_samples", "english.pbmm")
transcription_model = DeepSpeech(model_path)
audio_path = os.path.join("test_samples", "audio.wav")
transcription = transcription_model.transcribe(audio_path)
assert similarity(TEXT, transcription) > MIN_SYNTHESIS_SCORE
def sim_align(self, a, start, end):
source = self.text[start:end]
words = source.split(" ")
best = ""
best_score = 0
for i in range(len(words)):
for j in range(i, len(words)):
t = " ".join(words[i:j])
score = similarity(a, t)
if score > best_score:
best = t
best_score = score
start = self.text.index(best)
end = start + len(best)
return start, end, best_score
def test_hifigan_synthesis():
hifigan_model_path = os.path.join("test_samples", "hifigan.pt")
hifigan_config_path = os.path.join("test_samples", "config.json")
audio_path = "synthesized_audio.wav"
transcription_model = Silero()
hifigan = Hifigan(hifigan_model_path, hifigan_config_path)
text = "the monkeys live"
synthesize(
model=FakeModelForSynthesis(),
text=text,
graph_path=None,
audio_path=audio_path,
vocoder=hifigan,
)
assert os.path.isfile(audio_path)
assert similarity(
text, transcription_model.transcribe(audio_path)) > MIN_SYNTHESIS_SCORE
os.remove(audio_path)
def generate_clips_from_subtitles(
audio_path,
subs,
transcription_model,
output_path,
logging=logging,
min_length=MIN_LENGTH,
max_length=MAX_LENGTH,
min_confidence=MIN_CONFIDENCE,
):
"""
Generates clips from subtitles.
Parameters
----------
audio_path : str
Path to audio file (must have been converted using convert_audio)
subs : list
List of pysrt subtitle objects
transcription_model : TranscriptionModel
Transcription model
output_path : str
Path to save audio clips to
logging : logging (optional)
Logging object to write logs to
min_length : float (optional)
Minimum duration of a clip in seconds
max_length : float (optional)
Maximum duration of a clip in seconds
min_confidence : float (optional)
Minimum confidence score to generate a clip for
Returns
-------
(list, list)
List of clips and clip lengths in seconds
"""
logging.info("Loading subtitles...")
total = len(subs)
logging.info(f"{total} subtitle lines detected...")
result_fragments = []
unlabelled_fragments = []
clip_lengths = []
for i, sub in enumerate(subs):
duration = sub.duration.seconds + (sub.duration.milliseconds / 1000)
if duration >= min_length and duration <= max_length:
start = sub.start.to_time().strftime("%H:%M:%S.%f")
end = sub.end.to_time().strftime("%H:%M:%S.%f")
filename = cut_audio(audio_path, start, end, output_path)
clip_path = os.path.join(output_path, filename)
try:
transcript = transcription_model.transcribe(clip_path)
except:
logging.info(f"Could not transcribe {clip_path}")
transcript = None
if transcript:
text = sub.text.strip().replace("\n", " ")
score = similarity(transcript, text)
if score >= min_confidence:
result_fragments.append(
{
"name": filename,
"start": start,
"end": end,
"duration": duration,
"transcript": transcript,
"text": text,
"score": score,
}
)
clip_lengths.append(duration)
else:
unlabelled_fragments.append(filename)
logging.info(f"Progress - {i+1}/{total}")
return result_fragments, unlabelled_fragments, clip_lengths
def test_silero():
transcription_model = Silero()
audio_path = os.path.join("test_samples", "audio.wav")
transcription = transcription_model.transcribe(audio_path)
assert similarity(TEXT, transcription) > MIN_SYNTHESIS_SCORE
def test_similarity():
assert similarity("abc", "def") == 0
assert similarity("abc", "abc") == 1