app.py

import random 
import gradio as gr 
import numpy as np 
from elevenlabs import voices, generate, set_api_key, UnauthenticatedRateLimitError

def pad_buffer(audio):
    # Pad buffer to multiple of 2 bytes
    buffer_size = len(audio)
    element_size = np.dtype(np.int16).itemsize
    if buffer_size % element_size != 0:
        audio = audio + b'\0' * (element_size - (buffer_size % element_size))
    return audio 

def generate_voice(text, voice_name):
    model_name = "eleven_multilingual_v1"
    try:
        audio = generate(
            text[:250],  # Limit to 250 characters
            voice=voice_name, 
            model=model_name
        )
        return (44100, np.frombuffer(pad_buffer(audio), dtype=np.int16))
    except UnauthenticatedRateLimitError as e:
        raise gr.Error("Thanks for trying out ElevenLabs TTS! You've reached the free tier limit. Please provide an API key to continue.") 
    except Exception as e:
        raise gr.Error(str(e))

# Fetch all available voices
all_voices = voices()

# Filter out the desired voices and set Antonio as default
desired_voices = ["Anita", "Rachael", "Michael", "Antonio"]
filtered_voices = [voice.name for voice in all_voices if voice.name in desired_voices]

input_text = gr.Textbox(
    label="Input Text (250 characters max)", 
    lines=2, 
    value="Hahaha OHH MY GOD! This is SOOO funny, I-I am Eleven a text-to-speech system!",
    elem_id="input_text"
)

input_voice = gr.Dropdown(
    choices=filtered_voices, 
    default="Antonio",
    label="Voice", 
    elem_id="input_voice"
)

out_audio = gr.Audio(
    label="Generated Voice",
    type="numpy", 
    elem_id="out_audio"
)

iface = gr.Interface(
    fn=generate_voice,
    inputs=[input_text, input_voice],
    outputs=out_audio,
    live=True,
    theme="Monochrome",
    concurrency_count=1
)

iface.launch(debug=True)