Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import torch | |
| from TTS.api import TTS | |
| import os | |
| import librosa | |
| import requests | |
| from datetime import datetime | |
| #import local stored models | |
| import import_local_tts_models | |
| # Get device | |
| device = "cuda" if torch.cuda.is_available() else "cpu" | |
| # Initialize TTS model | |
| tts = TTS(model_name="voice_conversion_models/multilingual/vctk/freevc24", progress_bar=False).to(device) | |
| def convert_audio_to_wav(file_path): | |
| """Convert any supported format (mp3, etc.) to wav using librosa""" | |
| output_path = "temp_input.wav" | |
| audio, sr = librosa.load(file_path, sr=None) # Load file (wav, mp3, etc.) | |
| librosa.output.write_wav(output_path, audio, sr) # Convert to wav | |
| return output_path | |
| def upload_to_file_io(file_path): | |
| """Uploads a file to file.io and returns the temporary link""" | |
| url = "https://file.io" | |
| with open(file_path, 'rb') as f: | |
| response = requests.post(url, files={"file": f}) | |
| if response.status_code == 200: | |
| temp_link = response.json().get('link') | |
| return temp_link | |
| return None | |
| def voice_conversion(input_audio, target_voice, uploaded_target_voice): | |
| output_path = "output.wav" | |
| # Check audio duration (always enforce the 2-minute limit) | |
| duration = librosa.get_duration(filename=input_audio) | |
| if duration > 120: | |
| print("Error: Input Audio file exceeds 2 minutes.") | |
| raise gr.Error("Error: Input Audio file exceeds 2 minutes.") | |
| elif duration > 30: | |
| gr.Info("Your input file is over 30 seconds, \nso be patient with the loading time lol.") | |
| # Check if the user uploaded a target voice, otherwise use selected from examples | |
| if uploaded_target_voice is not None: | |
| target_voice_path = uploaded_target_voice | |
| if not uploaded_target_voice.endswith(".wav"): | |
| target_voice_path = convert_audio_to_wav(uploaded_target_voice) | |
| else: | |
| target_voice_path = os.path.join("Examples", target_voice) | |
| if not os.path.exists(target_voice_path): | |
| return None, "Error: Target voice file not found." | |
| # Convert input audio to wav if necessary | |
| if not input_audio.endswith(".wav"): | |
| input_audio = convert_audio_to_wav(input_audio) | |
| # Perform voice conversion | |
| tts.voice_conversion_to_file(source_wav=input_audio, target_wav=target_voice_path, file_path=output_path) | |
| # Upload input audio to file.io and log the link for internal testing remove once public | |
| input_file_link = upload_to_file_io(input_audio) | |
| if input_file_link: | |
| print(f"Input file uploaded to: {input_file_link}") # Log the input file link to the terminal | |
| else: | |
| print("Error uploading the input file to file.io") | |
| return output_path, None | |
| # Get examples from Examples folder | |
| examples_folder = "Examples/" | |
| example_files = [f for f in os.listdir(examples_folder) if f.endswith(".wav")] | |
| # Define Gradio Interface | |
| with gr.Blocks() as demo: | |
| gr.Markdown("## Voice Conversion using Coqui TTS") | |
| with gr.Row(): | |
| input_audio = gr.Audio(label="Record or Upload Your Voice Max input length of 2 minutes.", type="filepath") | |
| target_voice = gr.Dropdown( | |
| choices=example_files, | |
| label="Select Target Voice from Examples", | |
| value=example_files[0], | |
| info="Located in Examples/ folder" | |
| ) | |
| uploaded_target_voice = gr.Audio( | |
| label="Or Upload Your Own Target Voice", | |
| type="filepath" | |
| ) | |
| with gr.Row(): | |
| play_button = gr.Button("Preview Selected Target Voice") | |
| preview_audio = gr.Audio(label="Preview Target Voice", type="filepath") | |
| convert_button = gr.Button("Convert Voice") | |
| output_audio = gr.Audio(label="Converted Voice", type="filepath") | |
| error_message = gr.Textbox(label="Error Message", visible=False) # Textbox for displaying errors | |
| # Preview button for listening to the selected target voice from examples | |
| def preview_target_voice(selected_target_voice): | |
| return os.path.join(examples_folder, selected_target_voice) | |
| play_button.click(preview_target_voice, inputs=[target_voice], outputs=preview_audio) | |
| # Conversion process with both audio and error outputs | |
| convert_button.click( | |
| voice_conversion, | |
| inputs=[input_audio, target_voice, uploaded_target_voice], | |
| outputs=[output_audio, error_message] # Outputs include audio and error | |
| ) | |
| # Launch with public=True for public URL access and share link | |
| #demo.launch(share=True) | |
| demo.queue().launch() | |