bark_sample.py

import os
from bark import SAMPLE_RATE, generate_audio
from bark.generation import (
    preload_models
)
from scipy.io.wavfile import write as write_wav
import nltk
import numpy as np
import argparse

# Setup environment variables to speeds up generation
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
os.environ["SUNO_OFFLOAD_CPU"] = "1"
# os.environ["SUNO_USE_SMALL_MODELS"] = "1"

default_speaker = "v2/en_speaker_6"
test_script = """
"""
audio_file_extension = ".mp4"


def __generate_voice(script, speaker):
    sentences = nltk.sent_tokenize(script.replace("\n", " ").strip())
    silence = np.zeros(int(0.25 * SAMPLE_RATE))  # quarter second of silence
    pieces = []
    for sentence in sentences:
        print(f'\nGenerating audio for: "{sentence}"')
        audio_array = generate_audio(sentence, history_prompt=speaker)
        pieces += [audio_array, silence.copy()]
    return np.concatenate(pieces)


def english_gpt(script, speaker, output_file):
    # Download the models and sample data
    nltk.download('punkt')
    preload_models()
    audio_array = __generate_voice(script if script else test_script,
                                   speaker if speaker else default_speaker)
    default_output_file = test_script[0:35].replace('\n', '').strip()
    output_file = output_file if output_file else f'{default_output_file}.wav'
    write_wav(output_file, SAMPLE_RATE, audio_array)
    return output_file


def convert_to_mp4(wav_file, mp4_file):
    command = f"ffmpeg -i '{wav_file}' -c:v libx264 -c:a aac -strict -2 '{mp4_file}'"
    print(command)
    os.system(command)


def start_cli():
    parser = argparse.ArgumentParser(description='Generate audio from text')
    parser.add_argument('--script', help='Script to generate audio from')
    parser.add_argument('--input', help='Input file with script')
    parser.add_argument('--speaker', help='Speaker to use')
    parser.add_argument('--output', help='Output file')

    args = parser.parse_args()

    if args.input:
        with open(args.input, 'r') as file:
            my_script = file.read()
    else:
        my_script = args.script
    generated_audio = english_gpt(my_script, args.speaker, args.output)
    print(generated_audio)
    convert_to_mp4(generated_audio, os.path.splitext(generated_audio)[0] + audio_file_extension)


start_cli()