Option to run on CPU only

[tdryehthrehre]

Hi,

you can probably go through everything where it says cuda and change it to cpu, but an option to do this while installing would be very nice still. Or can you give a small guide what you have to change for it to run on CPU only?

Thanks.

[nonetrix]

AMD too pls

[0x41337]

@tdryehthrehre and @nonetrix

To support CPU you must edit the app.py file in:

if is_mac_os():
    device = torch.device('cpu')
else:
    device = torch.device('cuda:0')

then comment out these four lines and on line 27 define: device = torch.device('cpu') save the file and run it again. This way the torch engine will only use the CPU.

Here is the app.py with this patches:

app.py

import gradio as gr
import torch
import platform
import random
import json
from pathlib import Path
from TTS.api import TTS
import uuid
import html
import soundfile as sf

def is_mac_os():
    return platform.system() == 'Darwin'

params = {
    "activate": True,
    "autoplay": True,
    "show_text": False,
    "remove_trailing_dots": False,
    "voice": "Rogger.wav",
    "language": "English",
    "model_name": "tts_models/multilingual/multi-dataset/xtts_v2",
}

# SUPPORTED_FORMATS = ['wav', 'mp3', 'flac', 'ogg']
SAMPLE_RATE = 16000
device = torch.device('cpu')

# Set the default speaker name
default_speaker_name = "Rogger"

# if is_mac_os():
#     device = torch.device('cpu')
# else:
#     device = torch.device('cuda:0')

# Load model
tts = TTS(model_name=params["model_name"]).to(device)

# # Random sentence (assuming harvard_sentences.txt is in the correct path)
# def random_sentence():
#     with open(Path("harvard_sentences.txt")) as f:
#         return random.choice(list(f))

# Voice generation function
def gen_voice(string, spk, speed, english):
    string = html.unescape(string)
    short_uuid = str(uuid.uuid4())[:8]
    fl_name='outputs/' + spk + "-" + short_uuid +'.wav'
    output_file = Path(fl_name)
    this_dir = str(Path(__file__).parent.resolve())
    tts.tts_to_file(
        text=string,
        speed=speed,
        file_path=output_file,
        speaker_wav=[f"{this_dir}/targets/" +spk + ".wav"],
        language=languages[english]
    )
    return output_file

def update_speakers():
    speakers = {p.stem: str(p) for p in list(Path('targets').glob("*.wav"))}
    return list(speakers.keys())

def update_dropdown(_=None, selected_speaker=default_speaker_name):
    return gr.Dropdown(choices=update_speakers(), value=selected_speaker, label="Select Speaker")

def handle_recorded_audio(audio_data, speaker_dropdown, filename = "user_entered"):
    if not audio_data:
        return speaker_dropdown
    
    sample_rate, audio_content = audio_data
    
    save_path = f"targets/{filename}.wav"

    # Write the audio content to a WAV file
    sf.write(save_path, audio_content, sample_rate)

    # Create a new Dropdown with the updated speakers list, including the recorded audio
    updated_dropdown = update_dropdown(selected_speaker=filename)
    return updated_dropdown


# Load the language data
with open(Path('languages.json'), encoding='utf8') as f:
    languages = json.load(f)

# Gradio Blocks interface
with gr.Blocks() as app:
    
    gr.Markdown("### TTS based Voice Cloning.")
    
    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(lines=2, label="Speechify this Text",value="Even in the darkest nights, a single spark of hope can ignite the fire of determination within us, guiding us towards a future we dare to dream.")
            speed_slider = gr.Slider(label='Speed', minimum=0.1, maximum=1.99, value=0.8, step=0.01)
            language_dropdown = gr.Dropdown(list(languages.keys()), label="Language/Accent", value="English")

            gr.Markdown("### Speaker Selection and Voice Cloning")
            
            with gr.Row():
                with gr.Column():
                    speaker_dropdown = update_dropdown()
                    refresh_button = gr.Button("Refresh Speakers")
                with gr.Column():
                    filename_input = gr.Textbox(label="Add new Speaker", placeholder="Enter a name for your recording/upload to save as")
                    save_button = gr.Button("Save Below Recording")
                
            refresh_button.click(fn=update_dropdown, inputs=[], outputs=speaker_dropdown)

            with gr.Row():
                record_button = gr.Audio(label="Record Your Voice")
                
            save_button.click(fn=handle_recorded_audio, inputs=[record_button, speaker_dropdown, filename_input], outputs=speaker_dropdown)
            record_button.stop_recording(fn=handle_recorded_audio, inputs=[record_button, filename_input], outputs=speaker_dropdown)
            record_button.upload(fn=handle_recorded_audio, inputs=[record_button, filename_input], outputs=speaker_dropdown)
            
            submit_button = gr.Button("Convert")

        with gr.Column():
            audio_output = gr.Audio()

    submit_button.click(
        fn=gen_voice,
        inputs=[text_input, speaker_dropdown, speed_slider, language_dropdown],
        outputs=audio_output
    )

if __name__ == "__main__":
    app.launch()