import os, subprocess
import gradio as gr
import shutil, time, torch, gc
from mega import Mega
from datetime import datetime
import pandas as pd
import os, sys, subprocess, numpy as np
from pydub import AudioSegment
try:
from whisperspeech.pipeline import Pipeline as TTS
whisperspeak_on = True
except:
whisperspeak_on = False
# Class to handle caching model urls from a spreadsheet
class CachedModels:
def __init__(self):
csv_url = "https://docs.google.com/spreadsheets/d/1tAUaQrEHYgRsm1Lvrnj14HFHDwJWl0Bd9x0QePewNco/export?format=csv&gid=1977693859"
if os.path.exists("spreadsheet.csv"):
self.cached_data = pd.read_csv("spreadsheet.csv")
else:
self.cached_data = pd.read_csv(csv_url)
self.cached_data.to_csv("spreadsheet.csv", index=False)
# Cache model urls
self.models = {}
for _, row in self.cached_data.iterrows():
filename = row['Filename']
url = None
for value in row.values:
if isinstance(value, str) and "huggingface" in value:
url = value
break
if url:
self.models[filename] = url
# Get cached model urls
def get_models(self):
return self.models
def show(path,ext,on_error=None):
try:
return list(filter(lambda x: x.endswith(ext), os.listdir(path)))
except:
return on_error
def run_subprocess(command):
try:
subprocess.run(command, check=True)
return True, None
except Exception as e:
return False, e
def download_from_url(url=None, model=None):
if not url:
try:
url = model[f'{model}']
except:
gr.Warning("Failed")
return ''
if model == '':
try:
model = url.split('/')[-1].split('?')[0]
except:
gr.Warning('Please name the model')
return
model = model.replace('.pth', '').replace('.index', '').replace('.zip', '')
url = url.replace('/blob/main/', '/resolve/main/').strip()
for directory in ["downloads", "unzips","zip"]:
#shutil.rmtree(directory, ignore_errors=True)
os.makedirs(directory, exist_ok=True)
try:
if url.endswith('.pth'):
subprocess.run(["wget", url, "-O", f'assets/weights/{model}.pth'])
elif url.endswith('.index'):
os.makedirs(f'logs/{model}', exist_ok=True)
subprocess.run(["wget", url, "-O", f'logs/{model}/added_{model}.index'])
elif url.endswith('.zip'):
subprocess.run(["wget", url, "-O", f'downloads/{model}.zip'])
else:
if "drive.google.com" in url:
url = url.split('/')[0]
subprocess.run(["gdown", url, "--fuzzy", "-O", f'downloads/{model}'])
elif "mega.nz" in url:
Mega().download_url(url, 'downloads')
else:
subprocess.run(["wget", url, "-O", f'downloads/{model}'])
downloaded_file = next((f for f in os.listdir("downloads")), None)
if downloaded_file:
if downloaded_file.endswith(".zip"):
shutil.unpack_archive(f'downloads/{downloaded_file}', "unzips", 'zip')
for root, _, files in os.walk('unzips'):
for file in files:
file_path = os.path.join(root, file)
if file.endswith(".index"):
os.makedirs(f'logs/{model}', exist_ok=True)
shutil.copy2(file_path, f'logs/{model}')
elif file.endswith(".pth") and "G_" not in file and "D_" not in file:
shutil.copy(file_path, f'assets/weights/{model}.pth')
elif downloaded_file.endswith(".pth"):
shutil.copy(f'downloads/{downloaded_file}', f'assets/weights/{model}.pth')
elif downloaded_file.endswith(".index"):
os.makedirs(f'logs/{model}', exist_ok=True)
shutil.copy(f'downloads/{downloaded_file}', f'logs/{model}/added_{model}.index')
else:
gr.Warning("Failed to download file")
return 'Failed'
gr.Info("Done")
except Exception as e:
gr.Warning(f"There's been an error: {str(e)}")
finally:
shutil.rmtree("downloads", ignore_errors=True)
shutil.rmtree("unzips", ignore_errors=True)
shutil.rmtree("zip", ignore_errors=True)
return 'Done'
def speak(audio, text):
print(f"({audio}, {text})")
current_dir = os.getcwd()
os.chdir('./gpt_sovits_demo')
process = subprocess.Popen([
"python", "./zero.py",
"--input_file", audio,
"--audio_lang", "English",
"--text", text,
"--text_lang", "English"
], stdout=subprocess.PIPE, text=True)
for line in process.stdout:
line = line.strip()
if "All keys matched successfully" in line:
continue
if line.startswith("(") and line.endswith(")"):
path, finished = line[1:-1].split(", ")
if finished:
os.chdir(current_dir)
return path
os.chdir(current_dir)
return None
def whisperspeak(text, tts_lang, cps=10.5):
if whisperspeak_on is None: return None
if not "tts_pipe" in locals(): tts_pipe = TTS(t2s_ref='whisperspeech/whisperspeech:t2s-v1.95-small-8lang.model', s2a_ref='whisperspeech/whisperspeech:s2a-v1.95-medium-7lang.model')
from fastprogress.fastprogress import master_bar, progress_bar
master_bar.update = lambda *args, **kwargs: None
progress_bar.update = lambda *args, **kwargs: None
output = f"audios/tts_audio_{datetime.now().strftime('%Y%m%d_%H%M%S')}.wav"
tts_pipe.generate_to_file(output, text, cps=cps, lang=tts_lang)
return os.path.abspath(output)
def stereo_process(audio1,audio2,choice):
audio = audio1 if choice == "Input" else audio2
print(audio)
sample_rate, audio_array = audio
if len(audio_array.shape) == 1:
audio_bytes = audio_array.tobytes()
segment = AudioSegment(
data=audio_bytes,
sample_width=audio_array.dtype.itemsize, # 2 bytes for int16
frame_rate=sample_rate, # Use the sample rate from your tuple
channels=1 # Adjust if your audio has more channels
)
samples = np.array(segment.get_array_of_samples())
delay_samples = int(segment.frame_rate * (0.6 / 1000.0))
left_channel = np.zeros_like(samples)
right_channel = samples
left_channel[delay_samples:] = samples[:-delay_samples]
stereo_samples = np.column_stack((left_channel, right_channel))
return (sample_rate, stereo_samples.astype(np.int16))
else:
return audio
def sr_process(audio1, audio2, choice):
torch.cuda.empty_cache()
gc.collect()
if "tts_pipe" in locals(): del tts_pipe
audio = audio1 if choice == "Input" else audio2
sample_rate, audio_array = audio
audio_segment = AudioSegment(
audio_array.tobytes(),
frame_rate=sample_rate,
sample_width=audio_array.dtype.itemsize,
channels=1 if len(audio_array.shape) == 1 else 2
)
temp_file = os.path.join('TEMP', f'{choice}_{datetime.now().strftime("%Y%m%d_%H%M%S")}.wav')
audio_segment.export(temp_file, format="wav")
output_folder = "SR"
model_name = "speech"
suffix = "_ldm"
guidance_scale = 2.7
ddim_steps = 50
venv_dir = "audiosr"
def split_audio(input_file, output_folder, chunk_duration=5.12):
os.makedirs(output_folder, exist_ok=True)
ffmpeg_command = f"ffmpeg -i {input_file} -f segment -segment_time {chunk_duration} -c:a pcm_s16le {output_folder}/out%03d.wav"
subprocess.run(ffmpeg_command, shell=True, check=True)
def create_file_list(output_folder):
file_list = os.path.join(output_folder, "file_list.txt")
with open(file_list, "w") as f:
for filename in sorted(os.listdir(output_folder)):
if filename.endswith(".wav"):
f.write(os.path.join(output_folder, filename) + "\n")
return file_list
def run_audiosr(file_list, model_name, suffix, guidance_scale, ddim_steps, output_folder, venv_dir):
command = f"{venv_dir}/bin/python -m audiosr --input_file_list {file_list} --model_name {model_name} --suffix {suffix} --guidance_scale {guidance_scale} --ddim_steps {ddim_steps} --save_path {output_folder}"
try:
subprocess.run(command, shell=True, check=True, stderr=subprocess.PIPE)
except subprocess.CalledProcessError as e:
print(f"Error running audiosr: {e.stderr.decode()}")
split_audio(temp_file, output_folder)
file_list = create_file_list(output_folder)
run_audiosr(file_list, model_name, suffix, guidance_scale, ddim_steps, output_folder, venv_dir)
output_file = None
time.sleep(1)
processed_chunks = []
for root, dirs, files in os.walk(output_folder):
for file in sorted(files):
if file.startswith("out") and file.endswith(f"{suffix}.wav"):
chunk_file = os.path.join(root, file)
processed_chunks.append(AudioSegment.from_wav(chunk_file))
if processed_chunks:
merged_audio = sum(processed_chunks)
output_file = os.path.join(output_folder, f"{choice}_merged{suffix}.wav")
merged_audio.export(output_file, format="wav")
display_file = AudioSegment.from_file(output_file)
sample_rate = display_file.frame_rate
audio_array = np.array(display_file.get_array_of_samples())
return (sample_rate, audio_array)
else:
print(f"Error: Could not find any processed audio chunks in {output_folder}")
return None