database.py

import hashlib
import os
from glob import glob

import laion_clap
from diskcache import Cache
from qdrant_client import QdrantClient
from qdrant_client.http import models
from tqdm import tqdm


# Functions utils ##################################################################################
def get_md5(fpath):
    with open(fpath, "rb") as f:
        file_hash = hashlib.md5()
        while chunk := f.read(8192):
            file_hash.update(chunk)
    return file_hash.hexdigest()


# PARAMETERS #######################################################################################
CACHE_FOLDER = '/home/nahia/data/audio/'
KAGGLE_TRAIN_PATH = '/home/nahia/Documents/audio/actor/Actor_01/'

# ################## Loading the CLAP model ###################
print("[INFO] Loading the model...")
model_name = 'music_speech_epoch_15_esc_89.25.pt'
model = laion_clap.CLAP_Module(enable_fusion=False)
model.load_ckpt()  # download the default pretrained checkpoint.

# Initialize the cache
os.makedirs(CACHE_FOLDER, exist_ok=True)
cache = Cache(CACHE_FOLDER)

# Embed the audio files !
audio_files = [p for p in glob(os.path.join(KAGGLE_TRAIN_PATH, '*.wav'))]
audio_embeddings = []
chunk_size = 100
total_chunks = int(len(audio_files) / chunk_size)

# Use tqdm for a progress bar
for i in tqdm(range(0, len(audio_files), chunk_size), total=total_chunks):
    chunk = audio_files[i:i + chunk_size]  # Get a chunk of audio files
    chunk_embeddings = []

    for audio_file in chunk:
        # Compute a unique hash for the audio file
        file_key = get_md5(audio_file)

        if file_key in cache:
            # If the embedding for this file is cached, retrieve it
            embedding = cache[file_key]
        else:
            # Otherwise, compute the embedding and cache it
            embedding = model.get_audio_embedding_from_filelist(x=[audio_file], use_tensor=False)[
                0]  # Assuming the model returns a list
            cache[file_key] = embedding
        chunk_embeddings.append(embedding)
    audio_embeddings.extend(chunk_embeddings)

# It's a good practice to close the cache when done
cache.close()

# Creating a qdrant collection #####################################################################
client = QdrantClient("localhost", port=6333)
print("[INFO] Client created...")

print("[INFO] Creating qdrant data collection...")
client.create_collection(
     collection_name="demo_db7",
     vectors_config=models.VectorParams(
         size=audio_embeddings[0].shape[0],
         distance=models.Distance.COSINE
     ),
 )

# Créer des enregistrements Qdrant à partir des embeddings
records = []
for idx, (audio_path, embedding) in enumerate(zip(audio_files, audio_embeddings)):
    record = models.PointStruct(
        id=idx,
        vector=embedding,
        payload={"audio_path": audio_path, "style": audio_path.split('/')[-2]}
    )
    records.append(record)

# Téléverser les enregistrements dans la collection Qdrant
print("[INFO] Uploading data records to data collection...")
client.upload_points(collection_name="demo_db7", points=records)
print("[INFO] Successfully uploaded data records to data collection!")