app.py

import streamlit as st
from piano_transcription_inference import PianoTranscription, sample_rate, load_audio_from_memory
from io import BytesIO
import base64

st.title('Transcribe piano')

audiofile = st.file_uploader('Upload audio file', type=['.wav'], accept_multiple_files=False)
my_bar = None

def print_progress(current, total):
 my_bar.progress(current / total, text=f'Transcribing ({current + 1} / {total + 1} segments)...')

if audiofile is not None:
 audio_bytes = audiofile.read()
 st.text('Uploaded file')
 st.audio(audio_bytes, format='audio/wav')
 # (audio, _) = load_audio(audio_path, sr=sample_rate, mono=True)

 with st.spinner('Resampling...'):
 (audio, _) = load_audio_from_memory(audio_bytes, sr=sample_rate, mono=True)
 st.success('Resampling complete.')

 my_bar = st.progress(0, text='Transcribing...')

 transcriptor = PianoTranscription(device='cpu', checkpoint_path='CRNN_note_F1=0.9677_pedal_F1=0.9186.pth') # device: 'cuda' | 'cpu'

 buf = BytesIO()

 transcribed_dict = transcriptor.transcribe(audio, None, print_progress, buf)

 # st.download_button('Download MIDI', buf.getvalue(), f'transcribed_{audiofile.name}.mid', 'audio/midi')

 filename = f'transcribed_{audiofile.name}.mid'

 b64 = base64.b64encode(buf.getvalue()).decode()
 st.markdown(f'<a href="data:audio/midi;base64,{b64}" download="{filename}">Download MIDI</a>', unsafe_allow_html=True)
 st.balloons()