|
import streamlit as st |
|
from transformers import AutoProcessor, SeamlessM4Tv2Model |
|
import torchaudio |
|
import soundfile as sf |
|
import torch |
|
import os |
|
|
|
language_map = { |
|
"Modern Standard Arabic": "arb", "Bengali": "ben", "Catalan": "cat", |
|
"Czech": "ces", "Mandarin Chinese": "cmn", "Welsh": "cym", |
|
"Danish": "dan", "German": "deu", "English": "eng", |
|
"Estonian": "est", "Finnish": "fin", "French": "fra", |
|
"Hindi": "hin", "Indonesian": "ind", "Italian": "ita", |
|
"Japanese": "jpn", "Kannada": "kan", "Korean": "kor", |
|
"Maltese": "mlt", "Dutch": "nld", "Western Persian": "pes", |
|
"Polish": "pol", "Portuguese": "por", "Romanian": "ron", |
|
"Russian": "rus", "Slovak": "slk", "Spanish": "spa", |
|
"Swedish": "swe", "Swahili": "swh", "Tamil": "tam", |
|
"Telugu": "tel", "Tagalog": "tgl", "Thai": "tha", |
|
"Turkish": "tur", "Ukrainian": "ukr", "Urdu": "urd", |
|
"Northern Uzbek": "uzn", "Vietnamese": "vie" |
|
} |
|
|
|
|
|
device = "cuda" if torch.cuda.is_available() else "cpu" |
|
|
|
|
|
@st.cache_resource |
|
def load_model_and_processor(): |
|
processor = AutoProcessor.from_pretrained("facebook/seamless-m4t-v2-large") |
|
model = SeamlessM4Tv2Model.from_pretrained("facebook/seamless-m4t-v2-large") |
|
model.to(device) |
|
return processor, model |
|
|
|
processor, model = load_model_and_processor() |
|
|
|
|
|
st.title("Sarvm Audio Search") |
|
|
|
|
|
st.sidebar.header("Input Settings") |
|
input_text = st.sidebar.text_input("Enter text for conversion:", "Hello, my dog is cute") |
|
selected_language_name = st.sidebar.selectbox("Select Target Language:", list(language_map.keys())) |
|
selected_language_code = language_map[selected_language_name] |
|
|
|
|
|
def text_to_audio(text, language): |
|
text_inputs = processor(text=text, src_lang="eng", return_tensors="pt").to(device) |
|
audio_array = model.generate(**text_inputs, tgt_lang=language)[0].cpu().numpy().squeeze() |
|
file_path = 'audio_from_text.wav' |
|
sf.write(file_path, audio_array, 16000) |
|
return file_path |
|
|
|
|
|
|
|
def audio_to_audio(input_audio_path, language): |
|
audio, orig_freq = torchaudio.load(input_audio_path) |
|
audio = torchaudio.functional.resample(audio, orig_freq=orig_freq, new_freq=16000).to(device) |
|
audio_inputs = processor(audios=audio, return_tensors="pt").to(device) |
|
audio_array = model.generate(**audio_inputs, tgt_lang=language)[0].cpu().numpy().squeeze() |
|
file_path = 'audio_from_audio.wav' |
|
sf.write(file_path, audio_array, 16000) |
|
return file_path |
|
|
|
|
|
if st.sidebar.button("Convert Text to Audio"): |
|
with st.spinner("Converting..."): |
|
audio_path = text_to_audio(input_text, selected_language_code) |
|
st.audio(audio_path) |
|
|
|
|
|
uploaded_audio = st.sidebar.file_uploader("Upload audio for conversion:", type=["wav"]) |
|
if uploaded_audio is not None and st.button("Convert Uploaded Audio"): |
|
with st.spinner("Converting..."): |
|
audio_file_path = f"temp_{uploaded_audio.name}" |
|
with open(audio_file_path, "wb") as f: |
|
f.write(uploaded_audio.getvalue()) |
|
converted_audio_path = audio_to_audio(audio_file_path, selected_language_code) |
|
st.audio(converted_audio_path) |
|
os.remove(audio_file_path) |