Spaces:
Running
Running
import gradio as gr | |
from gradio_unifiedaudio import UnifiedAudio | |
from huggingface_hub import InferenceClient | |
from transformers import pipeline | |
import torch | |
import tempfile | |
import subprocess | |
import spaces | |
MODEL_NAME = "openai/whisper-large-v3" | |
BATCH_SIZE = 8 | |
FILE_LIMIT_MB = 1000 | |
YT_LENGTH_LIMIT_S = 3600 | |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") | |
# Initialize ASR pipeline | |
asr_pipe = pipeline( | |
task="automatic-speech-recognition", | |
model=MODEL_NAME, | |
chunk_length_s=30, | |
device=device, | |
) | |
# Initialize Mistral model client | |
client1 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1") | |
system_instructions1 = "[SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Tony Stark', Keep conversation very short, clear, friendly and concise. The text provided is a request for a specific type of response from you, the virtual assistant. The request asks you to provide friendly responses as if You are the character Jarvis, made by 'Tony Stark.' The expectation is that I will avoid introductions and start answering the query directly, elaborating on all aspects of the request. As an AI-powered assistant, your task is to generate responses that appear as if they are created by the user and not an AI-powered assistant. [USER]" | |
# History of interactions | |
history = [] | |
def transcribe(inputs, task="translate"): | |
if inputs is None: | |
raise ValueError("No audio file submitted! Please record an audio file before submitting your request.") | |
text = asr_pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)["text"] | |
return text | |
def model(text): | |
generate_kwargs = dict( | |
temperature=0.7, | |
max_new_tokens=512, | |
top_p=0.95, | |
repetition_penalty=1, | |
do_sample=True, | |
seed=42, | |
) | |
formatted_prompt = system_instructions1 + text + "[JARVIS]" | |
stream = client1.text_generation( | |
formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False) | |
output = "" | |
for response in stream: | |
if not response.token.text == "</s>": | |
output += response.token.text | |
return output | |
def process_audio(audio): | |
if not audio: | |
return "Please record an audio.", None | |
# Provide user feedback | |
print("Processing audio...") | |
text = transcribe(audio) | |
print("Whisper Response -> ", text) | |
history.append(text) | |
history_text = " ".join(history) | |
mistral_response = model(history_text) | |
print("Mistral Response -> ", mistral_response) | |
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file: | |
tmp_path = tmp_file.name | |
command = ["edge-tts", "-t", mistral_response, "--write-media", tmp_path] | |
print(' '.join(command)) | |
result = subprocess.run(command, capture_output=True, text=True) | |
if result.returncode == 0: | |
print("Command executed successfully.") | |
else: | |
print(f"Command failed with return code {result.returncode}") | |
print(f"Error message: {result.stderr}") | |
return mistral_response,tmp_path | |
def clear_audio(audio): | |
return UnifiedAudio(value=None),UnifiedAudio(value=None) | |
DESCRIPTION = """ # <center><b>Voice-Mistral-Voice ๐ค</b></center> | |
### <center>Voice-Mistral-Voice</center> | |
""" | |
MORE = """ ## TRY Other Models | |
### https://huggingface.co/spaces/KingNish/Instant-Video | |
### Instant Image: 4k images in 5 Second -> https://huggingface.co/spaces/KingNish/Instant-Image | |
""" | |
BETA = """ ### Voice Chat (BETA)""" | |
FAST = """## Fastest Model""" | |
with gr.Blocks() as demo: | |
gr.Markdown(DESCRIPTION) | |
with gr.Row(): | |
audio_input = UnifiedAudio(sources="microphone", type="filepath", image="./robot.png", container=True) | |
output_audio = UnifiedAudio(sources="microphone", type="filepath", image="./logo.png", autoplay=True) | |
audio_text = gr.Text() | |
gr.Markdown(FAST) | |
clear_audio_button = gr.Button("Record Again") | |
audio_input.change(process_audio, inputs=audio_input, outputs=[audio_text,output_audio]) | |
clear_audio_button.click(clear_audio, inputs=None, outputs=[audio_input, output_audio]) | |
gr.Markdown(MORE) | |
if __name__ == '__main__': | |
demo.queue(max_size=200).launch() | |