init

download_audio.py

@@ -84,9 +84,11 @@ def get_audio(dataframe: pd.DataFrame):
     for side, df in dataframe.groupby("side"):
         df.pop("side")
         features.update({f"{side}.{k}": to_json_serializable(v) for k, v in df.iloc[0].to_dict().items()})
-        features[f"{side}.path"] = str(p_join(cache_dir_audio, os.path.basename(features[f"{side}.url"])))
+        identifier = os.path.basename(features[f"{side}.url"]).split(".")[-1]
+        features[f"{side}.path"] = str(p_join(cache_dir_audio, side, f"{features['line_no']}.{identifier}"))
         if not os.path.exists(features[f"{side}.path"]):
-            if not wget(features[f"{side}.url"], output_file=features[f"{side}.path"]):
+            flag = wget(features[f"{side}.url"], output_file=features[f"{side}.path"])
+            if not flag:
                 return False
     with open(p_join(cache_dir_feature, f'{features["line_no"]}.json'), "w") as f:
         json.dump(features, f)

main.sh

@@ -2,7 +2,7 @@
 # enA-jaA: 718_606 #
 ####################
 export DIRECTION="enA-jaA"
-export MAX_RETRY=10
+
 export LINE_NO_START=0
 export LINE_NO_END=50000
 python download_audio.py

push_s2s_translation.py

@@ -0,0 +1,112 @@
+import json
+import os
+import time
+from os.path import join as p_join
+from tqdm import tqdm
+from typing import Dict
+from glob import glob
+
+from soundfile import LibsndfileError
+from datasets import Dataset, Audio, DatasetDict
+
+# dataset config
+direction = os.getenv("DIRECTION", "enA-jaA")
+sides = {i: n for n, i in enumerate(sorted(direction.split("-")), 1)}
+sides_rev = {v: k for k, v in sides.items()}
+cache_dir_audio = p_join("download", "audio", direction)
+cache_dir_feature = p_join("download", "feature", direction)
+os.makedirs(cache_dir_audio, exist_ok=True)
+os.makedirs(cache_dir_feature, exist_ok=True)
+line_no_start = int(os.getenv("LINE_NO_START", 0))
+line_no_end = int(os.getenv("LINE_NO_END", 10000))
+dataset_id = int(os.getenv("DATASET_ID", 0))
+hf_org = "kotoba-tech"
+hf_dataset = f"seamless-align-{direction}-{dataset_id}"
+
+
+def loader(feature: str) -> Dict:
+    with open(feature) as f:
+        return json.load(f)
+
+
+# create a dataset instance
+
+files = {
+    int(os.path.basename(i).replace(".json", "")): i for i in glob(p_join(cache_dir_feature, "*.json"))
+}
+file_ids = [i for i in range(line_no_start, line_no_end) if i in files]
+features = [loader(files[i]) for i in file_ids]
+print(f"features: {len(features)}")
+features = [i for i in features if os.path.exists(i[f"{sides_rev[1]}.path"]) and os.path.exists(i[f"{sides_rev[2]}.path"])]
+print(f"features (filtered): {len(features)}")
+data_dict = {
+    f"{sides_rev[1]}.audio": [i.pop(f"{sides_rev[1]}.path") for i in features],
+    f"{sides_rev[2]}.audio": [i.pop(f"{sides_rev[2]}.path") for i in features]
+}
+keys = features[0].keys()
+data_dict.update(
+    {k: [i[k] for i in features] for k in keys}
+)
+audio_dataset = Dataset.from_dict(data_dict)
+audio_dataset = audio_dataset.cast_column(f"{sides_rev[1]}.audio", Audio())
+audio_dataset = audio_dataset.cast_column(f"{sides_rev[2]}.audio", Audio())
+
+# remove instances with broken audio files
+broken_files = []
+for i in tqdm(range(len(audio_dataset))):
+    try:
+        a = audio_dataset[i]
+        flag = True
+        for side_id in sides_rev.keys():
+            start = a[f"{sides_rev[side_id]}.duration_start"]
+            end = a[f"{sides_rev[side_id]}.duration_end"]
+            array = a[f"{sides_rev[side_id]}.audio"]["array"]
+            flag = 0 < start < end < len(array)
+        if not flag:
+            broken_files.append(i)
+    except LibsndfileError:
+        broken_files.append(i)
+        continue
+print(f"features (removed broken audio): {len(audio_dataset) - len(broken_files)}")
+
+# remove broken files
+for i in broken_files:
+    if os.path.exists(files[file_ids[i]]):
+        os.remove(files[file_ids[i]])
+    for side_id in sides_rev.keys():
+        if os.path.exists(data_dict[f"{sides_rev[side_id]}.audio"][i]):
+            os.remove(data_dict[f"{sides_rev[side_id]}.audio"][i])
+valid_data_id = [i for i in range(len(audio_dataset)) if i not in broken_files]
+audio_dataset_valid = audio_dataset.select(valid_data_id)
+
+
+# trim the audio according to the duration
+def clip_audio(batch):
+    for side_id in sides_rev.keys():
+        start = batch[f"{sides_rev[side_id]}.duration_start"]
+        end = batch[f"{sides_rev[side_id]}.duration_end"]
+        audio = batch[f"{sides_rev[side_id]}.audio"]
+        batch[f"{sides_rev[side_id]}.audio"] = [
+            {"array": a["array"][s:e], "sampling_rate": a["sampling_rate"]}
+            for a, s, e in zip(audio, start, end)
+        ]
+    return batch
+
+
+audio_dataset_valid = audio_dataset_valid.map(
+    function=clip_audio,
+    batched=True,
+    batch_size=128,
+    num_proc=1,
+    desc="clipping audio based on the duration:"
+)
+
+dataset_to_push = DatasetDict({"train": audio_dataset_valid})
+repo_name = f"{hf_org}/{hf_dataset}"
+while True:
+    try:
+        dataset_to_push.push_to_hub(repo_name)
+        break
+    except Exception:
+        print(f"FAILED: push_to_hub on {repo_name} failed. wait 60 sec and retry soon...")
+        time.sleep(60)

requirements.txt

@@ -1,2 +1,5 @@
+datasets
+soundfile
+librosa
 requests
 pandas

upload_audio.py

@@ -1,31 +0,0 @@
-import json
-import os
-import tarfile
-import zipfile
-import gzip
-import subprocess
-from os.path import join as p_join
-from tqdm import tqdm
-from multiprocessing import Pool
-from typing import Optional
-
-import pandas as pd
-
-from datasets import Dataset, Audio
-
-# dataset config
-direction = os.getenv("DIRECTION", "enA-jaA")
-sides = set(direction.split("-"))
-cache_dir_audio = p_join("download", "audio", direction)
-cache_dir_feature = p_join("download", "feature", direction)
-os.makedirs(cache_dir_audio, exist_ok=True)
-os.makedirs(cache_dir_feature, exist_ok=True)
-# processor config
-n_pool = int(os.getenv("N_POOL", 8))
-wget_max_retry = os.getenv("MAX_RETRY", "1")
-wget_timeout = os.getenv("TIMEOUT", "20")
-line_no_start = int(os.getenv("LINE_NO_START", 0))
-line_no_end = int(os.getenv("LINE_NO_END", 10000))
-
-audio_dataset = Dataset.from_dict({"audio": ["path/to/audio_1", "path/to/audio_2", ..., "path/to/audio_n"]}).cast_column("audio", Audio())
-audio_dataset[0]["audio"]