import streamlit as st
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import pandas as pd
from fpdf import FPDF
import whisper
import tempfile
from st_audiorec import st_audiorec
import numpy as np
# Interface utilisateur
st.set_page_config(
page_title="Traduction de la parole en pictogrammes ARASAAC",
page_icon="đź“ť",
layout="wide"
)
# Charger le modèle et le tokenizer
checkpoint = "Propicto/t2p-nllb-200-distilled-600M-all"
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
# Charger le modèle Whisper
whisper_model = whisper.load_model("base")
# Lire le lexique
@st.cache_data
def read_lexicon(lexicon):
df = pd.read_csv(lexicon, sep='\t')
df['keyword_no_cat'] = df['lemma'].str.split(' #').str[0].str.strip().str.replace(' ', '_')
return df
lexicon = read_lexicon("lexicon.csv")
# Processus de sortie de la traduction
def process_output_trad(pred):
return pred.split()
def get_id_picto_from_predicted_lemma(df_lexicon, lemma):
if lemma.endswith("!"):
lemma = lemma[:-1]
id_picto = df_lexicon.loc[df_lexicon['keyword_no_cat'] == lemma, 'id_picto'].tolist()
return (id_picto[0], lemma) if id_picto else (0, lemma)
# Génération du contenu HTML pour afficher les pictogrammes
def generate_html(ids):
html_content = ''
for picto_id, lemma in ids:
if picto_id != 0: # ignore invalid IDs
img_url = f"https://static.arasaac.org/pictograms/{picto_id}/{picto_id}_500.png"
html_content += f'''
{lemma}
'''
html_content += ''
return html_content
# Génération du PDF
def generate_pdf(ids):
pdf = FPDF(orientation='L', unit='mm', format='A4') # 'L' for landscape orientation
pdf.add_page()
pdf.set_auto_page_break(auto=True, margin=15)
# Start positions
x_start = 10
y_start = 10
img_width = 50
img_height = 50
spacing = 1
max_width = 297 # A4 landscape width in mm
current_x = x_start
current_y = y_start
for picto_id, lemma in ids:
if picto_id != 0: # ignore invalid IDs
img_url = f"https://static.arasaac.org/pictograms/{picto_id}/{picto_id}_500.png"
pdf.image(img_url, x=current_x, y=current_y, w=img_width, h=img_height)
pdf.set_xy(current_x, current_y + img_height + 5)
pdf.set_font("Arial", size=12)
pdf.cell(img_width, 10, txt=lemma, ln=1, align='C')
current_x += img_width + spacing
# Move to the next line if exceeds max width
if current_x + img_width > max_width:
current_x = x_start
current_y += img_height + spacing + 10 # Adjust for image height and some spacing
pdf_path = "pictograms.pdf"
pdf.output(pdf_path)
return pdf_path
# Initialiser l'Ă©tat de session
if 'transcription' not in st.session_state:
st.session_state['transcription'] = None
if 'pictogram_ids' not in st.session_state:
st.session_state['pictogram_ids'] = None
if 'previous_audio_file' not in st.session_state:
st.session_state['previous_audio_file'] = None
# Interface utilisateur pour l'audio et le bouton de téléchargement
st.title("Traduction de la parole en pictogrammes ARASAAC")
col1, col2 = st.columns(2)
with col1:
audio_file = st.file_uploader("Ajouter un fichier audio :", type=["wav", "mp3"])
# RĂ©initialiser les informations si le fichier audio change
if audio_file is not None and audio_file != st.session_state['previous_audio_file']:
st.session_state['transcription'] = None
st.session_state['pictogram_ids'] = None
st.session_state['previous_audio_file'] = audio_file
with col2:
if audio_file is not None:
with st.spinner("Transcription de l'audio en cours..."):
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_file:
temp_file.write(audio_file.read())
temp_file_path = temp_file.name
transcription = whisper_model.transcribe(temp_file_path, language='fr')
if 'transcription' in locals():
st.text_area("Transcription :", transcription['text'])
st.session_state['transcription'] = transcription['text']
with st.spinner("Affichage des pictogrammes..."):
if st.session_state['transcription'] is not None:
inputs = tokenizer(transcription['text'].lower(), return_tensors="pt").input_ids
outputs = model.generate(inputs, max_new_tokens=40, do_sample=True, top_k=30, top_p=0.95)
pred = tokenizer.decode(outputs[0], skip_special_tokens=True)
sentence_to_map = process_output_trad(pred)
pictogram_ids = [get_id_picto_from_predicted_lemma(lexicon, lemma) for lemma in sentence_to_map]
st.session_state['pictogram_ids'] = [get_id_picto_from_predicted_lemma(lexicon, lemma) for lemma in sentence_to_map]
if st.session_state['pictogram_ids'] is not None:
html = generate_html(st.session_state['pictogram_ids'])
st.components.v1.html(html, height=500, scrolling=True)
# Container to hold the download button
pdf_path = generate_pdf(st.session_state['pictogram_ids'])
with open(pdf_path, "rb") as pdf_file:
st.download_button(label="Télécharger la traduction en PDF", data=pdf_file, file_name="pictograms.pdf", mime="application/pdf")
# record_audio = st_audiorec()
# if record_audio:
# audio = np.array(record_audio)
# transcription = whisper_model.transcribe(audio, language='fr')
# st.success("Enregistrement terminé !")