Paillat-dev/FABLE
1
0
Fork 0
mirror of https://github.com/Paillat-dev/FABLE.git synced 2026-01-02 01:06:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(audio_prompts): add default audio prompts for narrator

Browse Source 
feat(audio_prompts): add en_narrator_deep audio prompt for narrator
feat(audio_prompts): add en_narrator_light_bg audio prompt for narrator
fix(video.py): fix indentation and add prompt for generating thumbnail
fix(montage.py): fix indentation and add prompt for generating thumbnail
fix(montage.py): fix image download for wikimage slides

fix(speak.py): remove unused import statement
fix(speak.py): remove unused variable 'fakenames'
feat(speak.py): add function 'remove_blank_moments' to remove silent parts from audio file
feat(speak.py): add function 'optimize_string_groups' to optimize string groups for audio generation
fix(speak.py): fix comment indentation in 'generate_voice' function
fix(speak.py): remove unused imports in 'generate_voice' function
fix(speak.py): remove unused variable 'reduced_noise' in 'generate_voice' function
fix(speak.py): remove unused import statements in 'generate_voice' function
fix(speak.py): remove unused import statement for 'logging' module
fix(speak.py): remove unused print statements in 'main' function
fix(speak.py): remove unused import statement for 'logging' module
fix(speak.py): remove unused print statements in 'main' function
fix(speak.py):

fix(wiki_downloader.py): fix Google search URL to include correct query parameter
fix(wiki_downloader.py): reduce sleep time after page load to 1 second
fix(wiki_downloader.py): increase sleep time after image click to 5 seconds
This commit is contained in:
Paillat
2023-07-02 11:17:10 +02:00
parent f1de2ad596
commit f7835f6604
13 changed files with 206 additions and 114 deletions
Download Patch File Download Diff File Expand all files Collapse all files

85
generators/speak.py
View File

@@ -1,5 +1,5 @@
import os
from pydub import AudioSegment, silence
fakenames = {
"Alexander": "p230",
@@ -11,16 +11,70 @@ fakenames = {
voices = ["Alexander", "Benjamin", "Amelia", "Katherine", "Johanne"]
def remove_blank_moments(file_path, silence_thresh= -50, silence_chunk_len=500):
# Load audio file
audio = AudioSegment.from_wav(file_path)
# Detect non-silent parts
nonsilent_data = silence.detect_nonsilent(audio, min_silence_len=silence_chunk_len, silence_thresh=silence_thresh)
# Create new audio file
final_audio = AudioSegment.empty()
# Iterate over non-silent parts and append to the final_audio with 0.5 seconds before and after each segment
for idx, (start_i, end_i) in enumerate(nonsilent_data):
start_i = max(0, start_i - 500) # 0.5 seconds before
end_i += 500 # 0.5 seconds after
segment = audio[start_i:end_i]
# Only append silence after the first segment
if idx > 0:
final_audio += AudioSegment.silent(duration=500)
final_audio += segment
# Save the result
if not os.path.exists(os.path.abspath(os.path.join(os.getcwd(), "temp"))):
os.mkdir(os.path.abspath(os.path.join(os.getcwd(), "temp")))
tempfile_path = os.path.abspath(os.path.join(os.getcwd(), "temp", "temp.wav"))
final_audio.export(tempfile_path, format="wav")
os.remove(file_path)
os.rename(tempfile_path, file_path)
def optimize_string_groups(strings):
optimized_groups = []
current_group = []
current_length = 0
for string in strings:
string_length = len(string) + len(current_group) # Account for spaces between strings
if current_length + string_length <= 100:
current_group.append(string)
current_length += string_length
else:
optimized_groups.append(' '.join(current_group)) # Join strings with spaces
current_group = [string]
current_length = len(string)
if current_group:
optimized_groups.append(' '.join(current_group))
return optimized_groups
class VoiceGenerator:
def __init__(self, mode="Bark", speaker=""):
self.mode = mode
self.speaker = speaker
if mode == "Bark":
os.environ["XDG_CACHE_HOME"] = os.path.join(os.getcwd(), "bark_cache")
from bark import preload_models, generation
from bark import preload_models
print("Loading Bark voice generator")
preload_models()
self.speaker = "v2/en_speaker_6"
#self.speaker = os.path.abspath(os.path.join(os.getcwd(), "audio_prompts", "en_male_professional_reader.npz"))
self.speaker = os.path.join(os.getcwd(), "audio_prompts", "en_narrator_light_bg.npz")
print(f"Generating voice for Bark with speaker {self.speaker}")
else:
from TTS.api import TTS
model = "tts_models/en/vctk/vits"
@@ -43,20 +97,27 @@ class VoiceGenerator:
import numpy as np
import nltk
sentences = nltk.sent_tokenize(text)
sentences = optimize_string_groups(sentences)
print(sentences)
pieces = []
silence = np.zeros(int(0.25 * SAMPLE_RATE)) # quarter second of silence
for sentence in sentences:
audio_array = generate_audio(sentence, history_prompt=self.speaker)
pieces += [audio_array, silence.copy()]
if not sentence == "":
audio_array = generate_audio(sentence, history_prompt=self.speaker)
pieces += [audio_array, silence.copy()]
audio_array = np.concatenate(pieces)
soundfile.write(path, audio_array, SAMPLE_RATE, format="WAV", subtype="PCM_16")
rate, data = wavread(path)
reduced_noise = nr.reduce_noise(y=data, sr=rate)
os.remove(path)
wavwrite(path, rate, reduced_noise)
'''
remove silence
'''
remove_blank_moments(path)
else:
self.tts.tts_to_file(text=text, file_path=path, speaker=self.speaker, speed=1, emotion="Happy")
if __name__ == "__main__":
import logging
logging.basicConfig(level=logging.INFO)
print("Testing voice generator")
generator = VoiceGenerator()
generator.generate_voice("test/test_r.wav", "Hello there!")
generator.generate_voice("test/teste_r.wav", "This is a test. I like the words python, django and flask. Betty bought a bit of butter but the butter was bitter. So she bought some better butter to make the bitter butter better.")
print("Loaded voice generator")
# generator.generate_voice("test/test_r.wav", "Hello there!")
generator.generate_voice("test/tast_timbernerslee.wav", "But his greatest claim to fame is undoubtedly his invention of the World Wide Web back in 1989. Can you imagine a world without the internet? [Laughs] No, thank you!")