app.py

import os
import io
from typing_extensions import Literal

import gradio as gr
import tempfile
import numpy as np

from dotenv import load_dotenv
from elevenlabs.client import ElevenLabs
from elevenlabs import play, stream, save
from elevenlabs import Voice, VoiceSettings
from pydub import AudioSegment
from pydub.playback import play
import imageio_ffmpeg as ffmpeg
import requests
from audiostretchy.stretch import AudioStretch


load_dotenv()

def verify_auth(username, password):
    if username == USER and password == PASSWORD:

        return True
    else:
        return False

ELE_API_KEY = os.getenv("ELE_API_KEY")
USER = os.getenv("USER")
PASSWORD = os.getenv("PASSWORD")

MODEL = "eleven_multilingual_v2"

ele_client = ElevenLabs(api_key=ELE_API_KEY)

VOICE = [
    "승현",
    "우승"
]

KEY_MAPPING = {
    "승현": "0RBbbgk6KUJxHmWzPiHz", # 승현+제시카(2:1)
    "우승": "ASwOiisDbuaP2R1jUQU6", # 우승+TTS_KKC(1:1)
}


AudioSegment.converter = ffmpeg.get_ffmpeg_exe()


def change_pitch(audio_segment, pitch_shift):
    new_sample_rate = int(audio_segment.frame_rate * (2.0 ** pitch_shift))
    pitched_audio = audio_segment._spawn(audio_segment.raw_data, overrides={'frame_rate': new_sample_rate})    
    return pitched_audio.set_frame_rate(audio_segment.frame_rate)


def predict(
        text: str,
        voice: str,
        output_file_format: Literal["mp3"] = "",
        speed: float = 1.0,
        pitch_shift: float = 0.0,
        stability: float = 0.5,
        similarity: float = 0.7,
        style_exaggeration: float = 0.,
        speaker_boost: bool = True     
):
    
    try:                        
        voice_setup=Voice(
            voice_id=KEY_MAPPING[voice],
            settings=VoiceSettings(stability=stability, similarity_boost=similarity, style=style_exaggeration, use_speaker_boost=speaker_boost)
        )
        
        audio = ele_client.generate(
            text = text,
            voice = voice_setup,
            model=MODEL
        )            
        audio_data = b''.join(audio)

    except Exception as e:
        raise requests.exceptions.RequestException(f"An error occurred while generating speech. Please check your API key and come back try again. {str(e)}")   
        
    print(f"[Text] {text}")
    
    audio_stretch = AudioStretch()
    audio_stretch.open_mp3(io.BytesIO(audio_data))
    audio_stretch.stretch(ratio=1/speed) # speed 0.5 -> 2.0\

    # Export the final audio to a temporary file
    with tempfile.NamedTemporaryFile(suffix=f".{output_file_format}", delete=False) as temp_file:
        audio_stretch.save(path=temp_file.name)
        audio = AudioSegment.from_file(temp_file.name)
        
        # Adjust pitch if needed
        if pitch_shift != 0.0:
            audio = change_pitch(audio, pitch_shift)
        
        audio.export(temp_file.name, format=output_file_format)
        temp_file_path = temp_file.name

    return temp_file_path

          
with gr.Blocks() as demo:
    gr.Markdown("# <center> Letsur Text-To-Speech API with Gradio </center>")
    with gr.Row(variant="panel"):        
        voice = gr.Dropdown(choices=VOICE, label="Voice Options", value="승현")
        output_file_format = gr.Dropdown(choices=["mp3"], label="Output Options", value="mp3")

    text = gr.Textbox(label="Input text",
                      value="안녕하세요.",
                      placeholder="안녕하세요.")
    
    # Additional parameters
    with gr.Accordion("Advanced Settings", open=False):
        speed = gr.Slider(label="speed", minimum=0.8, maximum=1.2, step=0.1, value=1.0)
        pitch_shift = gr.Slider(label="pitch_shift", minimum=-0.1, maximum=0.1, step=0.05, value=0.0) # 범위: 0~1
        stability = gr.Slider(label="stability", minimum=0., maximum=1., step=0.1, value=1.0) # 범위: 0~1
        similarity = gr.Slider(label="similarity", minimum=0., maximum=1., step=0.1, value=1.0) # 범위: 0~1
        style_exaggeration = gr.Slider(label="style_exaggeration", minimum=0., maximum=1., step=0.1, value=0.) # 범위: 0~1
        speaker_boost = gr.Checkbox(label="speaker_boost", value=True) # True or False            
    
    btn = gr.Button("Text-To-Speech")
    output_audio = gr.Audio(label="Speech Output")

    inputs = [text, voice, output_file_format] + [speed, pitch_shift, stability, similarity, style_exaggeration, speaker_boost]
    
    text.submit(fn=predict, inputs=inputs, outputs=output_audio, api_name="predict")
    btn.click(fn=predict, inputs=inputs, outputs=output_audio, api_name=False)

demo.queue().launch()