OuteTTS-DEMO / app.py
drewThomasson's picture
Now the user doesn't have to wait for the que to know if their input was too long :)
3271f83 verified
raw
history blame
3.48 kB
import gradio as gr
from outetts.v0_1.interface import InterfaceHF
import torch
# Initialize the TTS model interface
interface = InterfaceHF("OuteAI/OuteTTS-0.1-350M")
# Check if running on a CPU
is_cpu = not torch.cuda.is_available()
# Define a function to generate and save TTS output from input text
def generate_tts(text, temperature=0.1, repetition_penalty=1.1, max_length=4096):
# Set a character limit for the text input
max_characters = 30 # adjust as needed
# Check if input text exceeds character limit when on CPU
if is_cpu and len(text) > max_characters:
raise gr.Error(
f"Text input is too long! Please limit to {max_characters} characters.\nThis limit is in place to prevent long processing times as this interface is running on a free CPU tier."
)
# Log user input and parameters in the terminal
print(f"User entered text: {text}")
print(f"Temperature set to: {temperature}")
print(f"Repetition Penalty set to: {repetition_penalty}")
print(f"Max Length set to: {max_length}")
# Generate TTS output
output = interface.generate(
text=text,
temperature=temperature,
repetition_penalty=repetition_penalty,
max_lenght=max_length
)
# Save the output audio to a file
output.save("output.wav")
print("Audio generated and saved as output.wav")
return "output.wav"
# Create the Gradio Blocks interface
with gr.Blocks() as demo:
# Log each interaction
def on_text_input(text):
print(f"User typed text: {text}")
def on_temperature_change(val):
print(f"Temperature slider adjusted to: {val}")
def on_repetition_penalty_change(val):
print(f"Repetition Penalty slider adjusted to: {val}")
def on_max_length_change(val):
print(f"Max Length slider adjusted to: {val}")
# Dynamically set max_chars for text input based on whether it's CPU or GPU
if is_cpu:
text_input = gr.Textbox(
lines=2,
placeholder="Enter text to convert to speech (30 character limit on CPU)",
label="Text",
max_length=30 # Enforce character limit only on CPU
)
else:
text_input = gr.Textbox(
lines=2,
placeholder="Enter text to convert to speech",
label="Text"
)
# Track changes for debugging
text_input.change(on_text_input, inputs=text_input)
# Sliders with change events for tracking
temperature_slider = gr.Slider(0.1, 1.0, value=0.1, label="Temperature")
temperature_slider.change(on_temperature_change, inputs=temperature_slider)
repetition_penalty_slider = gr.Slider(1.0, 2.0, value=1.1, label="Repetition Penalty")
repetition_penalty_slider.change(on_repetition_penalty_change, inputs=repetition_penalty_slider)
max_length_slider = gr.Slider(512, 4096, value=4096, step=256, label="Max Length")
max_length_slider.change(on_max_length_change, inputs=max_length_slider)
# Button to generate TTS and Audio output
generate_button = gr.Button("Generate Speech")
audio_output = gr.Audio(type="filepath", label="Generated Speech")
# Define interaction between input and output
generate_button.click(
generate_tts,
inputs=[text_input, temperature_slider, repetition_penalty_slider, max_length_slider],
outputs=audio_output
)
print("Launching Gradio interface...")
demo.launch()