Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import streamlit as st
|
| 2 |
+
import os
|
| 3 |
+
import soundfile as sf
|
| 4 |
+
import torch
|
| 5 |
+
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
|
| 6 |
+
# Assuming you have your .env file configured with necessary API keys or configurations
|
| 7 |
+
# load_dotenv()
|
| 8 |
+
|
| 9 |
+
# Initialize the model outside the main app function to load it only once
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
| 13 |
+
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
| 14 |
+
|
| 15 |
+
model_id = "distil-whisper/distil-large-v2"
|
| 16 |
+
|
| 17 |
+
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
| 18 |
+
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
|
| 19 |
+
)
|
| 20 |
+
model.to(device)
|
| 21 |
+
|
| 22 |
+
processor = AutoProcessor.from_pretrained(model_id)
|
| 23 |
+
|
| 24 |
+
pipe = pipeline(
|
| 25 |
+
"automatic-speech-recognition",
|
| 26 |
+
model=model,
|
| 27 |
+
tokenizer=processor.tokenizer,
|
| 28 |
+
feature_extractor=processor.feature_extractor,
|
| 29 |
+
max_new_tokens=128,
|
| 30 |
+
chunk_length_s=15,
|
| 31 |
+
batch_size=16,
|
| 32 |
+
torch_dtype=torch_dtype,
|
| 33 |
+
device=device,
|
| 34 |
+
)
|
| 35 |
+
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
def transcribe_audio(audio_file):
|
| 39 |
+
# Save the audio file to a temporary file
|
| 40 |
+
with open("temp_audio_file", "wb") as f:
|
| 41 |
+
f.write(audio_file.getbuffer())
|
| 42 |
+
|
| 43 |
+
# Transcribe the audio file using the Whisper model
|
| 44 |
+
result = pipe("temp_audio_file")
|
| 45 |
+
return result["text"]
|
| 46 |
+
|
| 47 |
+
# Streamlit app
|
| 48 |
+
def main():
|
| 49 |
+
st.title('BETTER TRANSCRIBER')
|
| 50 |
+
|
| 51 |
+
# Audio file uploader
|
| 52 |
+
uploaded_file = st.file_uploader("Upload an audio file", type=["wav", "mp3", "m4a", "ogg", "flac"])
|
| 53 |
+
|
| 54 |
+
if uploaded_file is not None:
|
| 55 |
+
# Show a button to start the transcription process
|
| 56 |
+
if st.button('Transcribe'):
|
| 57 |
+
# Show a message while transcribing
|
| 58 |
+
with st.spinner('Transcribing...'):
|
| 59 |
+
text = transcribe_audio(uploaded_file)
|
| 60 |
+
|
| 61 |
+
# Show the transcription
|
| 62 |
+
st.subheader('Transcription:')
|
| 63 |
+
st.write(text)
|
| 64 |
+
else:
|
| 65 |
+
st.write('Upload an audio file to get started.')
|
| 66 |
+
|
| 67 |
+
if __name__ == "__main__":
|
| 68 |
+
main()
|