app.py

import gradio as gr
import os
import bark
from bark import generate_audio, preload_models, SAMPLE_RATE
import time
import scipy
import noisereduce as nr
import bark
from transformers import BertTokenizer

########################
##### Voice cloning functionality

# make sure to only use CPU
os.environ["CUDA_VISIBLE_DEVICES"] = ""
os.environ["SUNO_USE_SMALL_MODELS"] = "1"

# make sure to download BERT tokenizer
BertTokenizer.from_pretrained("bert-base-multilingual-cased")

# Do not re-download the models when loading them
bark.generation.CACHE_DIR = "bark_models"

def generate_cloned_voice_audio(text_prompt):

    print("="*10)
    print("NOW READING:")
    print(text_prompt)
    print("="*10)
    
    # load voice file
    history_prompt = "pm_voice.npz"

    # keep track of duration 
    t0 = time.time()

    # generate cloned voice audio
    audio_array = generate_audio(
        text_prompt,
        history_prompt = history_prompt
    )

    # keep track of duration 
    generation_duration_s = time.time() - t0
    audio_duration_s = audio_array.shape[0] / SAMPLE_RATE
    print(f"took {generation_duration_s:.0f}s to generate {audio_duration_s:.0f}s of audio")

    # reduce noise
    reduced_noise_audio_array = nr.reduce_noise(y=audio_array, sr=SAMPLE_RATE)
    
    # write to file
    audio_output_path = "output_audio.wav"
    noisereduced_audio_output_path = "output_noisereduced_audio.wav"
    scipy.io.wavfile.write(audio_output_path, rate=SAMPLE_RATE, data=audio_array)
    scipy.io.wavfile.write(noisereduced_audio_output_path, rate=SAMPLE_RATE, data=reduced_noise_audio_array)

    return (SAMPLE_RATE, audio_array)

########################

def greet(name):
    if os.path.isfile("pm_voice.npz"):
        preffix = "Found the voice file"
    else:
        preffix = "Voice file not found"
    return "Hello " + name + "!!" + preffix


output_audio = gr.Audio(
#    format = "ogg",
    label = "My cloned voice reading your text",   
)

iface = gr.Interface(
    fn=generate_cloned_voice_audio, 
    inputs="text", 
    outputs=output_audio
    )
iface.launch(share=True)