app.py

import gradio as gr
import torch
import torchaudio
from agc import AGC
import tempfile
import numpy as np
import lz4.frame
import os
from typing import Generator
import spaces

# Attempt to use GPU, fallback to CPU
try:
    torch_device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    print(f"Using device: {torch_device}")
except Exception as e:
    print(f"Error detecting GPU. Using CPU. Error: {e}")
    torch_device = torch.device("cpu")

# Load the AGC model
@spaces.GPU(duration=180)
def load_agc_model():
    return AGC.from_pretrained("Audiogen/agc-continuous").to(torch_device)

agc = load_agc_model()

@spaces.GPU(duration=180)
def encode_audio(audio_file_path):
    try:
        # Load the audio file
        waveform, sample_rate = torchaudio.load(audio_file_path)

        # Resample to 32kHz if necessary
        if sample_rate != 32000:
            resampler = torchaudio.transforms.Resample(sample_rate, 32000)
            waveform = resampler(waveform)

        # Convert to 32 channels if necessary
        if waveform.size(0) < 32:
            waveform = waveform.repeat(32, 1)[:32, :]
        elif waveform.size(0) > 32:
            waveform = waveform[:32, :]

        # Encode the audio
        audio = waveform.unsqueeze(0).to(torch_device)
        with torch.no_grad():
            z = agc.encode(audio)

        # Convert to NumPy and save to a temporary .owie file
        z_numpy = z.detach().cpu().numpy()
        temp_fd, temp_file_path = tempfile.mkstemp(suffix=".owie")
        os.close(temp_fd)  # Close the file descriptor to avoid issues with os.fdopen
        with open(temp_file_path, 'wb') as temp_file:
            compressed_data = lz4.frame.compress(z_numpy.tobytes())
            temp_file.write(compressed_data)
        
        return temp_file_path

    except Exception as e:
        return f"Encoding error: {e}"

@spaces.GPU(duration=180)
def decode_audio(encoded_file_path):
    try:
        # Load encoded data from the .owie file
        with open(encoded_file_path, 'rb') as temp_file:
            compressed_data = temp_file.read()
            z_numpy_bytes = lz4.frame.decompress(compressed_data)
            z_numpy = np.frombuffer(z_numpy_bytes, dtype=np.float32).reshape(1, 32, -1)
            z = torch.from_numpy(z_numpy).to(torch_device)

        # Decode the audio
        with torch.no_grad():
            reconstructed_audio = agc.decode(z)

        # Save to a temporary WAV file
        temp_wav_path = tempfile.mktemp(suffix=".wav")
        torchaudio.save(temp_wav_path, reconstructed_audio.squeeze(0).cpu(), 32000)
        return temp_wav_path

    except Exception as e:
        return f"Decoding error: {e}"

@spaces.GPU(duration=180)
def stream_decode_audio(encoded_file_path) -> Generator[tuple, None, None]:
    try:
        # Load encoded data from the .owie file
        with open(encoded_file_path, 'rb') as temp_file:
            compressed_data = temp_file.read()
            z_numpy_bytes = lz4.frame.decompress(compressed_data)
            z_numpy = np.frombuffer(z_numpy_bytes, dtype=np.float32).reshape(1, 32, -1)
            z = torch.from_numpy(z_numpy).to(torch_device)

        # Decode the audio in chunks
        chunk_size = 32000  # 1 second of audio at 32kHz
        sample_rate = 32000  # AGC model's output sample rate
        with torch.no_grad():
            for i in range(0, z.shape[2], chunk_size):
                z_chunk = z[:, :, i:i+chunk_size]
                audio_chunk = agc.decode(z_chunk)
                # Convert to numpy array (32 channels)
                audio_data = audio_chunk.squeeze(0).cpu().numpy()
                yield (sample_rate, audio_data)

    except Exception as e:
        print(f"Streaming decoding error: {e}")
        yield (sample_rate, np.zeros((32, chunk_size), dtype=np.float32))  # Return silence in case of error

# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("## Audio Compression with AGC (GPU/CPU) - 32 channels, 32kHz")

    with gr.Tab("Encode"):
        input_audio = gr.Audio(label="Input Audio", type="filepath")
        encode_button = gr.Button("Encode")
        encoded_output = gr.File(label="Encoded File (.owie)", type="filepath")

        encode_button.click(encode_audio, inputs=input_audio, outputs=encoded_output)

    with gr.Tab("Decode"):
        input_encoded = gr.File(label="Encoded File (.owie)", type="filepath")
        decode_button = gr.Button("Decode")
        decoded_output = gr.Audio(label="Decoded Audio", type="filepath")

        decode_button.click(decode_audio, inputs=input_encoded, outputs=decoded_output)

    with gr.Tab("Streaming"):
        input_encoded_stream = gr.File(label="Encoded File (.owie)", type="filepath")
        stream_button = gr.Button("Start Streaming")
        audio_output = gr.Audio(label="Streaming Audio Output", streaming=True)

        stream_button.click(stream_decode_audio, inputs=input_encoded_stream, outputs=audio_output)

demo.queue().launch()