import json import gradio as gr from gradio import Dropdown import os import shutil import re user_choice = "" tts = TTS("tts_models/multilingual/multi-dataset/bark", gpu=True) def infer(prompt, input_wav_file, clean_audio, hidden_numpy_audio): print(""" ————— NEW INFERENCE: ——————— """) if prompt == "": gr.Warning("Do not forget to provide a tts prompt !") if clean_audio is True: print("We want to clean audio sample") new_name = os.path.splitext(os.path.basename(input_wav_file))[0] if os.path.exists(os.path.join("bark_voices", f"{new_name}_cleaned")): print("This file has already been cleaned") check_name = os.path.join("bark_voices", f"{new_name}_cleaned") source_path = os.path.join(check_name, f"{new_name}_cleaned.wav") else: source_path = split_process(hidden_numpy_audio, "vocals") new_path = os.path.join(os.path.dirname( source_path), f"{new_name}_cleaned.wav") os.rename(source_path, new_path) source_path = new_path else: source_path = input_wav_file destination_directory = "bark_voices" file_name = os.path.splitext(os.path.basename(source_path))[0] destination_path = os.path.join(destination_directory, file_name) os.makedirs(destination_path, exist_ok=True) shutil.move(source_path, os.path.join( destination_path, f"{file_name}.wav")) sentences = re.split(r'(?<=[.!?])\s+', prompt) if len(sentences) > MAX_NUMBER_SENTENCES: gr.Info("Your text is too long. To keep this demo enjoyable for everyone, we only kept the first 10 sentences :) Duplicate this space and set MAX_NUMBER_SENTENCES for longer texts ;)") first_nb_sentences = sentences[:MAX_NUMBER_SENTENCES] limited_prompt = ' '.join(first_nb_sentences) prompt = limited_prompt else: prompt = prompt gr.Info("Generating audio from prompt") tts.tts_to_file(text=prompt, file_path="output.wav", voice_dir="bark_voices/", speaker=f"{file_name}") contents = os.listdir(f"bark_voices/{file_name}") for item in contents: print(item) print("Preparing final waveform video ...") tts_video = gr.make_waveform(audio="output.wav") print(tts_video) print("FINISHED") return "output.wav", tts_video, gr.update(value=f"bark_voices/{file_name}/{contents[1]}", visible=True), gr.Group.update(visible=True), destination_path prompt_choices = [ "I am very displeased with the progress being made to finish the cross-town transit line. transit line. This has been an embarrassing use of taxpayer dollars.", "Yes, John is my friend, but He was never at my house watching the baseball game.", "We are expecting a double digit increase in profits by the end of the fiscal year.", "Hi Grandma, Just calling to ask for money, or I can't see you over the holidays. " ] positive_prompts = { prompt_choices[0]: "I am very pleased with the progress being made to finish the cross-town transit line. This has been an excellent use of taxpayer dollars.", prompt_choices[1]: "Yes, John is my friend. He was at my house watching the baseball game all night.", prompt_choices[2]: "We are expecting a modest single digit increase in profits by the end of the fiscal year.", prompt_choices[3]: "Hi Grandma it’s me, Just calling to say I love you, and I can’t wait to see you over the holidays." } prompt = Dropdown( label="Text to speech prompt", choices=prompt_choices, elem_id="tts-prompt" ) css = """ #col-container {max-width: 780px; margin-left: auto; margin-right: auto;} a {text-decoration-line: underline; font-weight: 600;} .mic-wrap > button { width: 100%; height: 60px; font-size: 1.4em!important; } .record-icon.svelte-1thnwz { display: flex; position: relative; margin-right: var(--size-2); width: unset; height: unset; } span.record-icon > span.dot.svelte-1thnwz { width: 20px!important; height: 20px!important; } .animate-spin { animation: spin 1s linear infinite; } @keyframes spin { from { transform: rotate(0deg); } to { transform: rotate(360deg); } } """ def load_hidden_mic(audio_in): print("USER RECORDED A NEW SAMPLE") return audio_in def update_positive_prompt(prompt_value): global user_choice user_choice = prompt_value if prompt_value in positive_prompts: return positive_prompts[prompt_value] with gr.Blocks(css=css) as demo: with gr.Column(elem_id="col-container"): with gr.Row(): with gr.Column(): prompt = gr.Dropdown( label="Negative Speech Prompt", choices=prompt_choices, elem_id="tts-prompt" ) texts_samples = gr.Textbox( label="Positive prompts", info="Please read out this prompt 5 times to generate a good sample", value="", lines=5, elem_id="texts_samples" ) # Connect the prompt change to the update_positive_prompt function prompt.change(fn=update_positive_prompt, inputs=prompt, outputs=texts_samples) # Replace file input with microphone input micro_in = gr.Audio( label="Record voice to clone", type="filepath", source="microphone", interactive=True ) hidden_audio_numpy = gr.Audio(type="numpy", visible=False) submit_btn = gr.Button("Submit") with gr.Column(): cloned_out = gr.Audio( label="Text to speech output", visible=False) video_out = gr.Video(label="Waveform video", elem_id="voice-video-out") npz_file = gr.File(label=".npz file", visible=False) folder_path = gr.Textbox(visible=False) micro_in.stop_recording(fn=load_hidden_mic, inputs=[micro_in], outputs=[ hidden_audio_numpy], queue=False) submit_btn.click( fn=infer, inputs=[ prompt, micro_in, hidden_audio_numpy ], outputs=[ cloned_out, video_out, npz_file, folder_path ] ) demo.queue(api_open=False, max_size=10).launch()