Generate Audio from Text Scripts using Self-Hosted Bark Model and Google Drive

Audio Generator – Documentation

🎯 Purpose: Generate audio files from text scripts stored in Google Drive.

🔁 Flow:

1. Receive repo IDs.
2. Fetch text scripts.
3. Generate .wav files using local Bark model.
4. Upload back to Drive.

📦 Dependencies:

• Python script: /scripts/generate_voice.py
• Bark (voice generation system)
• n8n instance with access to local shell
• Google Drive OAuth2 credentials

✏️ Notes:

• Script filenames must end with .txt
• Only works with plain text
• No external API used = 100% free

📦 /scripts/generate_voice.py:

import sys
import torch
import numpy
import re
from bark import SAMPLE_RATE, generate_audio, preload_models
from scipy.io.wavfile import write as write_wav

# Patch to allow numpy._core.multiarray.scalar during loading
torch.serialization.add_safe_globals([numpy._core.multiarray.scalar])

# Monkey patch torch.load to force weights_only=False
_original_torch_load = torch.load
def patched_torch_load(f, *args, **kwargs):
    if 'weights_only' not in kwargs:
        kwargs['weights_only'] = False
    return _original_torch_load(f, *args, **kwargs)
torch.load = patched_torch_load

# Preload Bark models
preload_models()

def split_text(text, max_len=300):
    # Split on punctuation to avoid mid-sentence cuts
    sentences = re.split(r'(?<=[.?!])\s+', text)
    chunks = []
    current = ""
    for sentence in sentences:
        if len(current) + len(sentence) < max_len:
            current += sentence + " "
        else:
            chunks.append(current.strip())
            current = sentence + " "
    if current:
        chunks.append(current.strip())
    return chunks

# Input text file and output path
input_text_path = sys.argv[1]
output_wav_path = sys.argv[2]

with open(input_text_path, 'r', encoding='utf-8') as f:
    full_text = f.read()

voice_preset = "v2/en_speaker_7"

chunks = split_text(full_text)

# Generate and concatenate audio chunks
audio_arrays = []
for chunk in chunks:
    print(f"Generating audio for chunk: {chunk[:50]}...")
    audio = generate_audio(chunk, history_prompt=voice_preset)
    audio_arrays.append(audio)

# Merge all audio chunks
final_audio = numpy.concatenate(audio_arrays)

# Write final .wav file
write_wav(output_wav_path, SAMPLE_RATE, final_audio)

print(f"Full audio generated at: {output_wav_path}")