OPENGIT/Retrieval-based-Voice-Conversion-WebUI
1
0
Fork 0
mirror of https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI.git synced 2025-04-04 03:38:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add text splitting

Browse Source
This commit is contained in:
VSlobolinskyi 2025-03-21 18:30:57 +02:00
parent 9a0af2de15
commit a18a2c6b94
2 changed files with 154 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
modules/merged_ui/main.py
View File

@ -4,7 +4,6 @@ import gradio as gr
from merged_ui.utils import generate_and_process_with_rvc, modified_get_vc
from rvc_ui.initialization import config
from rvc_ui.main import names, index_paths
from spark.sparktts.utils.token_parser import LEVELS_MAP_UI
def build_merged_ui():
"""
@ -18,6 +17,7 @@ def build_merged_ui():
with gr.Tabs():
with gr.TabItem("TTS-to-RVC Pipeline"):
gr.Markdown("### Generate speech with Spark TTS and convert with RVC")
gr.Markdown("*Note: For multi-sentence text, each sentence will be processed separately and then combined.*")
# TTS Generation Section
with gr.Row():
@ -36,7 +36,7 @@ def build_merged_ui():
tts_text_input = gr.Textbox(
label="Text to synthesize",
lines=3,
placeholder="Enter text for TTS"
placeholder="Enter text for TTS. Multiple sentences will be processed individually."
)
prompt_text_input = gr.Textbox(
label="Text of prompt speech (Optional)",
@ -130,8 +130,8 @@ def build_merged_ui():
generate_with_rvc_button = gr.Button("Generate with RVC", variant="primary")
with gr.Row():
vc_output1 = gr.Textbox(label="Output information")
vc_output2 = gr.Audio(label="Final converted audio")
vc_output1 = gr.Textbox(label="Output information", lines=10)
vc_output2 = gr.Audio(label="Final concatenated audio")
# Connect generate function to button
generate_with_rvc_button.click(

176
modules/merged_ui/utils.py
View File

@ -1,5 +1,10 @@
import os
import shutil
import re
import numpy as np
from time import sleep
import soundfile as sf
from pydub import AudioSegment
# Import modules from your packages
from rvc_ui.initialization import vc
@ -11,31 +16,45 @@ model_dir = "spark/pretrained_models/Spark-TTS-0.5B"
device = 0
spark_model = initialize_model(model_dir, device=device)
def generate_and_process_with_rvc(
text, prompt_text, prompt_wav_upload, prompt_wav_record,
def split_into_sentences(text):
"""
Split text into sentences using regular expressions.
Args:
text (str): The input text to split
Returns:
list: A list of sentences
"""
# Split on period, exclamation mark, or question mark followed by space or end of string
sentences = re.split(r'(?<=[.!?])\s+|(?<=[.!?])$', text)
# Remove any empty sentences
sentences = [s.strip() for s in sentences if s.strip()]
return sentences
def process_single_sentence(
sentence_index, sentence, prompt_speech, prompt_text_clean,
spk_item, vc_transform, f0method,
file_index1, file_index2, index_rate, filter_radius,
resample_sr, rms_mix_rate, protect
resample_sr, rms_mix_rate, protect,
base_fragment_num
):
"""
Handle combined TTS and RVC processing and save outputs to TEMP directories
Process a single sentence through the TTS and RVC pipeline
Args:
sentence_index (int): Index of the sentence in the original text
sentence (str): The sentence text to process
... (other parameters are the same as generate_and_process_with_rvc)
Returns:
tuple: (spark_output_path, rvc_output_path, success, info_message)
"""
# Ensure TEMP directories exist
os.makedirs("./TEMP/spark", exist_ok=True)
os.makedirs("./TEMP/rvc", exist_ok=True)
# Get next fragment number
fragment_num = 1
while (os.path.exists(f"./TEMP/spark/fragment_{fragment_num}.wav") or
os.path.exists(f"./TEMP/rvc/fragment_{fragment_num}.wav")):
fragment_num += 1
# First generate TTS audio
prompt_speech = prompt_wav_upload if prompt_wav_upload else prompt_wav_record
prompt_text_clean = None if not prompt_text or len(prompt_text) < 2 else prompt_text
fragment_num = base_fragment_num + sentence_index
# Generate TTS audio for this sentence
tts_path = run_tts(
text,
sentence,
spark_model,
prompt_text=prompt_text_clean,
prompt_speech=prompt_speech
@ -43,7 +62,7 @@ def generate_and_process_with_rvc(
# Make sure we have a TTS file to process
if not tts_path or not os.path.exists(tts_path):
return "Failed to generate TTS audio", None
return None, None, False, f"Failed to generate TTS audio for sentence: {sentence}"
# Save Spark output to TEMP/spark
spark_output_path = f"./TEMP/spark/fragment_{fragment_num}.wav"
@ -70,7 +89,6 @@ def generate_and_process_with_rvc(
elif isinstance(output_audio, tuple) and len(output_audio) >= 2:
# Case 2: output_audio might be (sample_rate, audio_data)
try:
import soundfile as sf
sf.write(rvc_output_path, output_audio[1], output_audio[0])
rvc_saved = True
except Exception as inner_e:
@ -82,14 +100,120 @@ def generate_and_process_with_rvc(
except Exception as e:
output_info += f"\nError saving RVC output: {str(e)}"
# Add file paths to output info
output_info += f"\nSpark output saved to: {spark_output_path}"
# Prepare info message
info_message = f"Sentence {sentence_index+1}: {sentence[:30]}{'...' if len(sentence) > 30 else ''}\n"
info_message += f" - Spark output: {spark_output_path}\n"
if rvc_saved:
output_info += f"\nRVC output saved to: {rvc_output_path}"
info_message += f" - RVC output: {rvc_output_path}"
else:
output_info += f"\nCould not automatically save RVC output to {rvc_output_path}"
info_message += f" - Could not save RVC output to {rvc_output_path}"
return output_info, output_audio
return spark_output_path, rvc_output_path, rvc_saved, info_message
def concatenate_audio_files(file_paths, output_path, sample_rate=44100):
"""
Concatenate multiple audio files into a single file
Args:
file_paths (list): List of paths to audio files
output_path (str): Path to save the concatenated audio
sample_rate (int): Sample rate for the output file
Returns:
bool: True if concatenation was successful, False otherwise
"""
try:
# Use pydub to concatenate audio files
combined = AudioSegment.empty()
for file_path in file_paths:
segment = AudioSegment.from_file(file_path)
combined += segment
# Export the combined audio
combined.export(output_path, format="wav")
return True
except Exception as e:
print(f"Error concatenating audio files: {str(e)}")
# Fallback method using soundfile
try:
all_audio = []
for file_path in file_paths:
data, sr = sf.read(file_path)
# Convert to mono if stereo
if len(data.shape) > 1 and data.shape[1] > 1:
data = data.mean(axis=1)
all_audio.append(data)
# Concatenate all audio data
concatenated = np.concatenate(all_audio)
sf.write(output_path, concatenated, sample_rate)
return True
except Exception as e2:
print(f"Fallback concatenation failed: {str(e2)}")
return False
def generate_and_process_with_rvc(
text, prompt_text, prompt_wav_upload, prompt_wav_record,
spk_item, vc_transform, f0method,
file_index1, file_index2, index_rate, filter_radius,
resample_sr, rms_mix_rate, protect
):
"""
Handle combined TTS and RVC processing for multiple sentences and save outputs to TEMP directories
"""
# Ensure TEMP directories exist
os.makedirs("./TEMP/spark", exist_ok=True)
os.makedirs("./TEMP/rvc", exist_ok=True)
# Split text into sentences
sentences = split_into_sentences(text)
if not sentences:
return "No valid text to process.", None
# Get next base fragment number
base_fragment_num = 1
while any(os.path.exists(f"./TEMP/spark/fragment_{base_fragment_num + i}.wav") or
os.path.exists(f"./TEMP/rvc/fragment_{base_fragment_num + i}.wav")
for i in range(len(sentences))):
base_fragment_num += 1
# Process reference speech
prompt_speech = prompt_wav_upload if prompt_wav_upload else prompt_wav_record
prompt_text_clean = None if not prompt_text or len(prompt_text) < 2 else prompt_text
# Process each sentence
results = []
info_messages = [f"Processing {len(sentences)} sentences..."]
for i, sentence in enumerate(sentences):
spark_path, rvc_path, success, info = process_single_sentence(
i, sentence, prompt_speech, prompt_text_clean,
spk_item, vc_transform, f0method,
file_index1, file_index2, index_rate, filter_radius,
resample_sr, rms_mix_rate, protect,
base_fragment_num
)
info_messages.append(info)
if success and rvc_path:
results.append(rvc_path)
# If no sentences were successfully processed
if not results:
return "\n".join(info_messages) + "\n\nNo sentences were successfully processed.", None
# Concatenate all successful RVC fragments
final_output_path = f"./TEMP/final_output_{base_fragment_num}.wav"
concatenation_success = concatenate_audio_files(results, final_output_path)
if concatenation_success:
info_messages.append(f"\nAll fragments concatenated successfully to: {final_output_path}")
return "\n".join(info_messages), final_output_path
else:
# If concatenation failed but we have at least one successful fragment, return the first one
info_messages.append(f"\nFailed to concatenate fragments. Returning first successful fragment.")
return "\n".join(info_messages), results[0]
def modified_get_vc(sid0_value, protect0_value, file_index2_component):
"""
@ -101,4 +225,4 @@ def modified_get_vc(sid0_value, protect0_value, file_index2_component):
if isinstance(outputs, tuple) and len(outputs) >= 3:
return outputs[0], outputs[1], outputs[3]
return 0, protect0_value, file_index2_component.choices[0] if file_index2_component.choices else ""
return 0, protect0_value, file_index2_component.choices[0] if file_index2_component.choices else ""