app.py

import gradio as gr

import os
import wave
import json
import requests
import urllib.request
import speech_recognition
from vosk_tts import Model, Synth
from scipy.io.wavfile import write
from pydub import AudioSegment
from pydub.playback import play


PATH_TO_MODEL = os.path.join(os.getcwd(), "vosk-model-tts-ru-0.4-multi")
PATH_TO_OUTPUT = os.path.join(os.getcwd(), "content")
k = "sk-YOVNQzHmpga9My3dwlSo9BQN907TuPZQXcHn50ztigTwm3I2"
YA_GPT_KEY = "AQVNyVqBi-XoJ1cAo7VIxq6ztgXm3owqowtso5Qb"
instruction = """
Ответь на запрос так, как ответил бы на него Павел Воля. Используй данные из биографии Павла Воли, если это потребуется. Отвечай на запрос в его стиле. Ответ должен содержать не болеее 10 предложений.
"""


#files = [
#    ("input_face", open("C:\\Users\\user\\Desktop\\deepfake_sirius\\materials\\scale_1200.jpg", "rb")), #TODO: IT
#    ("input_audio", open("C:\\Users\\user\\Desktop\\deepfake_sirius\\materials\\audio\\output.wav", "rb")), #TODO: IT
#]


#     while True:
#         # старт записи речи с последующим выводом распознанной речи
#         # и удалением записанного в микрофон аудио
#         voice_input = record_and_recognize_audio()
#         os.remove("microphone-results.wav")
#         print(voice_input)
#         path_to_file = vg.generate(ask(voice_input))
#         print(path_to_file)
#         response = requests.post(
#             "https://api.gooey.ai/v2/Lipsync/form/",
#             headers={
#                 "Authorization": "Bearer " + k,
#             },
#             files=files,
#             data={"json": json.dumps(payload)},
#         )
#         assert response.ok, response.content
#         #song = AudioSegment.from_wav(path_to_file)
#         result = response.json()
#         print(response.status_code, result["output"]["output_video"])
#         #play(song)
#         urllib.request.urlretrieve(result["output"]["output_video"], "C:\\Users\\user\\Desktop\\deepfake_sirius\\materials\\video.mp4")
#         os.startfile("C:\\Users\\user\\Desktop\\deepfake_sirius\\materials\\video.mp4")
#         break;


class VoiceGenerator:
    
    def __init__(self):
        self.model = Model(PATH_TO_MODEL)

        
    def generate(self, text, file_name='output.wav'):
        synth = Synth(self.model)
        path = os.path.join(PATH_TO_OUTPUT, file_name)
        synth.synth(text, path)
        return path


def recognize_audio(file_path):
    with speech_recognition.AudioFile(file_path) as source:
        audio = recognizer.record(source)
    try:
        recognized_data = recognizer.recognize_google(audio, language="ru").lower()
    except speech_recognition.UnknownValueError:
        pass
    except speech_recognition.RequestError:
        pass
    return recognized_data


def ask_gpt(request):
    result = requests.post(
        url='https://llm.api.cloud.yandex.net/llm/v1alpha/instruct',
        headers={
            "Authorization": "Api-Key " + YA_GPT_KEY,
        },
        json={
            "model": "general",
            "instruction_text": instruction,
            "request_text": request,
            "generation_options": {
                "max_tokens": 1500,
                "temperature": 0.5
            }
        }
    )
    data = json.loads(result.text)
    return(data['result']['alternatives'][0]['text'])


recognizer = speech_recognition.Recognizer()
vg = VoiceGenerator()


def get_video(pathToWav, pathToImage):
    files = [
       ("input_face", open(pathToImage, "rb")),
       ("input_audio", open(pathToWav, "rb")),
    ]
    payload = {}
    response = requests.post(
        "https://api.gooey.ai/v2/Lipsync/form/",
        headers={
            "Authorization": "Bearer " + k,
        },
        files = files,
        data={"json": json.dumps(payload)}
    )
    assert response.ok, response.content


def result(audio):
    sample_rate, data = audio
    #print('sample_rate:', sample_rate, 'data:', data)
    #return os.path.join(os.path.abspath(''), "video_sample.mp4")
    file_name = 'voice_input.wav'
    file_path = os.path.join(PATH_TO_OUTPUT, file_name)
    write(file_path, sample_rate, data)
    text_from_audio = recognize_audio(file_path)
    generated_audio = vg.generate(ask_gpt(text_from_audio))
    referenceName = ''
    pathToReference = os.path.join(os.getcwd(), os.path.join("content", referenceName))
    return generated_audio


demo = gr.Interface(
    result,
    gr.Audio(sources=["microphone"]),
    "audio", #playable_video
)

demo.launch()