Spaces:
Running
Running
| import streamlit as st | |
| import edge_tts | |
| import asyncio | |
| import tempfile | |
| import os | |
| from huggingface_hub import InferenceClient | |
| import re | |
| from streaming_stt_nemo import Model | |
| import torch | |
| import random | |
| default_lang = "en" | |
| engines = {default_lang: Model(default_lang)} | |
| def transcribe(audio): | |
| lang = "en" | |
| model = engines[lang] | |
| text = model.stt_file(audio)[0] | |
| return text | |
| HF_TOKEN = os.environ.get("HF_TOKEN", None) | |
| def randomize_seed_fn(seed: int) -> int: | |
| seed = random.randint(0, 999999) | |
| return seed | |
| system_instructions1 = """ | |
| [SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Tony Stark.' | |
| Keep conversation friendly, short, clear, and concise. | |
| Avoid unnecessary introductions and answer the user's questions directly. | |
| Respond in a normal, conversational manner while being friendly and helpful. | |
| [USER] | |
| """ | |
| def models(text, seed=42): | |
| seed = int(randomize_seed_fn(seed)) | |
| generator = torch.Generator().manual_seed(seed) | |
| client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3") | |
| generate_kwargs = dict( | |
| max_new_tokens=300, | |
| seed=seed | |
| ) | |
| formatted_prompt = system_instructions1 + text + "[JARVIS]" | |
| stream = client.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 | |
| async def respond(audio, model, seed): | |
| user = transcribe(audio) | |
| reply = models(user, model, seed) | |
| communicate = edge_tts.Communicate(reply) | |
| with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file: | |
| tmp_path = tmp_file.name | |
| await communicate.save(tmp_path) | |
| return tmp_path | |
| DESCRIPTION = """ # <center><b>JARVIS⚡</b></center> | |
| ### <center>A personal Assistant of Tony Stark for YOU | |
| ### <center>Voice Chat with your personal Assistant</center> | |
| """ | |
| st.markdown(DESCRIPTION) | |
| st.title("JARVIS") | |
| uploaded_file = st.file_uploader("Upload audio file", type=["wav"]) | |
| seed = st.slider("Seed", min_value=0, max_value=999999, value=0) | |
| if uploaded_file is not None: | |
| # Convert the uploaded file to a BytesIO object | |
| audio_bytes = uploaded_file.read() | |
| # Process the audio using the respond function | |
| response_path = asyncio.run(respond(audio_bytes, models, seed)) | |
| # Display the audio response | |
| st.audio(response_path, format="audio/wav") | |
| os.remove(response_path) |