quivr/loaders/audio.py
2023-05-13 01:12:51 +02:00

56 lines
2.2 KiB
Python

import os
import tempfile
from io import BytesIO
import time
import openai
import streamlit as st
from langchain.document_loaders import TextLoader
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from utils import compute_sha1_from_content
from langchain.schema import Document
# Create a function to transcribe audio using Whisper
def _transcribe_audio(api_key, audio_file):
openai.api_key = api_key
transcript = ""
with BytesIO(audio_file.read()) as audio_bytes:
# Get the extension of the uploaded file
file_extension = os.path.splitext(audio_file.name)[-1]
# Create a temporary file with the uploaded audio data and the correct extension
with tempfile.NamedTemporaryFile(delete=True, suffix=file_extension) as temp_audio_file:
temp_audio_file.write(audio_bytes.read())
temp_audio_file.seek(0) # Move the file pointer to the beginning of the file
# Transcribe the temporary audio file
transcript = openai.Audio.translate("whisper-1", temp_audio_file)
return transcript
def process_audio(vector_store, file_name):
file_sha = ""
dateshort = time.strftime("%Y%m%d-%H%M%S")
file_meta_name = f"audiotranscript_{dateshort}.txt"
openai_api_key = st.secrets["openai_api_key"]
transcript = _transcribe_audio(openai_api_key, file_name)
file_sha = compute_sha1_from_content(transcript.text.encode("utf-8"))
## file size computed from transcript
file_size = len(transcript.text.encode("utf-8"))
## Load chunk size and overlap from sidebar
chunk_size = st.session_state['chunk_size']
chunk_overlap = st.session_state['chunk_overlap']
text_splitter = RecursiveCharacterTextSplitter.from_tiktoken_encoder(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
texts = text_splitter.split_text(transcript.text)
docs_with_metadata = [Document(page_content=text, metadata={"file_sha1": file_sha,"file_size": file_size, "file_name": file_meta_name, "chunk_size": chunk_size, "chunk_overlap": chunk_overlap, "date": dateshort}) for text in texts]
vector_store.add_documents(docs_with_metadata)
return vector_store