116_FreesoundLoopDataset/sft.py

import json
import tqdm
import re
import os
from pydub import AudioSegment

ann_path = "/work/fast_data_yinghao/116_FreesoundLoopDataset/sft_data-116_FreesoundLoopDataset-genre-classification.jsonl"
with open(ann_path, "r") as file:
    tmp1 = file.read()
    tmp1 = [ eval(tmp) for tmp in tmp1.split("\n")[:-1]]
    tmp1 = [i for i in tmp1 if i["output"] != "[]" and i["input"] != "[] \n"]
for data in tmp1:
    data["output"] = ",".join(eval(data["output"]))
    data["input"] = re.sub("\n","",data["input"])
    data["audioid"] = data["input"].split("<|SOA|>")[1].split("<|EOA|>")[0][25:] + ".wav"
    data["split"] = [data["split"]]

ann_path = "/work/fast_data_yinghao/116_FreesoundLoopDataset/sft_data-116_FreesoundLoopDataset-key_detection.jsonl"
with open(ann_path, "r") as file:
    tmp2 = file.read()
    tmp2 = [ eval(tmp) for tmp in tmp2.split("\n")[:-1]]
for data in tmp2:
    if data["output"] == "A":
        data["output"] = data["input"].split("A:")[1].split("\n")[0][9:]
    elif data["output"] == "B":
        data["output"] = data["input"].split("B:")[1].split("\n")[0][9:]
    data["output"] = re.sub("#","\u266f",data["output"])
    data["audioid"] = data["input"].split("<|SOA|>")[1].split("<|EOA|>")[0][25:] + ".wav.wav"
    file_path = os.path.join(f"/work/fast_data_yinghao/116_FreesoundLoopDataset/audio/wav", data["audioid"])
    if not os.path.exists(file_path):
        data["audioid"] = data["audioid"][:-8] + ".aiff.wav"
    data["input"] = data["input"].split("option")[0].split("dio:")[1]
    data["split"] = [data["split"]]

data_samples = tmp1 + tmp2
new_data_samples = []
train, valid, test = [], [], []
for data in tqdm.tqdm(data_samples):
    file_path = os.path.join(f"/work/fast_data_yinghao/116_FreesoundLoopDataset/audio/wav", data["audioid"])
    audio = AudioSegment.from_file(file_path)
    if len(audio) < 5000:
        continue
    if data["split"][0] == "train":
        train.append(data)
    elif data["split"][0] == "dev":
        valid.append(data)
    else:
        test.append(data)

for split in ["train", "valid", "test"]:
    with open(f"FSLD_{split}.jsonl", 'w') as outfile:
        for sample in eval(split):
            json.dump(sample, outfile)

            outfile.write('\n')
    outfile.close()

# print(f"Data successfully transformed and saved to {output_file_path}")