[dataloader] Fix text filtering bug and speed up spectrum length calc

examples/aishell-3/run.sh

@@ -40,6 +40,13 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     $dataset_dir/data_aishell3 \
     $data/all.txt
 
+  # Compute spec length (optional, but recommended)
+  python tools/compute_spec_length.py \
+    $data/all.txt \
+    $config \
+    $data/all_spec_length.txt
+  mv $data/all_spec_length.txt $data/all.txt
+
   cat $data/all.txt | awk -F '|' '{print $2}' | \
     sort | uniq | awk '{print $0, NR-1}' > $data/speaker.txt
   echo 'sil 0' > $data/phones.txt

requirements.txt

@@ -11,3 +11,4 @@ torchvision
 tqdm
 transformers
 huggingface_hub
+soundfile

tools/compute_spec_length.py

@@ -0,0 +1,72 @@
+#!/usr/bin/env python3
+# author: @lsrami
+
+import os
+import sys
+import json
+from tqdm import tqdm
+import soundfile as sf
+from concurrent.futures import ThreadPoolExecutor
+
+
+def load_filepaths_and_text(filename, split="|"):
+    with open(filename, encoding="utf-8") as f:
+        filepaths_and_text = [line.strip().split(split) for line in f]
+    return filepaths_and_text
+
+
+def process_item(item):
+    audiopath = item[0]
+    src_sampling_rate = sf.info(audiopath).samplerate
+    text = item[2]
+    text = text.strip().split()
+    if min_text_len <= len(text) and len(text) <= max_text_len:
+        length = int(os.path.getsize(audiopath) * sampling_rate /
+                     src_sampling_rate) // (2 * hop_length)
+        item.append(length)
+        return item
+    else:
+        return None
+
+
+def main(in_file, out_file):
+    """
+    Filter text & store spec lengths
+    """
+
+    audiopaths_sid_text = load_filepaths_and_text(in_file, split="|")
+
+    with ThreadPoolExecutor(max_workers=32) as executor:
+        results = list(
+            tqdm(
+                executor.map(process_item, audiopaths_sid_text),
+                total=len(audiopaths_sid_text),
+            )
+        )
+
+    # Filter out None results
+    results = [result for result in results if result is not None]
+
+    with open(out_file, "w", encoding="utf-8") as f:
+        for item in results:
+            f.write("|".join([str(i) for i in item]) + "\n")
+
+
+if __name__ == "__main__":
+    if len(sys.argv) != 4:
+        print(f"Usage: {sys.argv[0]} <in_file> <config_file> <out_file>")
+        sys.exit(1)
+    in_file, config_file, out_file = sys.argv[1:4]
+
+    with open(config_file, "r", encoding="utf8") as f:
+        data = f.read()
+    config = json.loads(data)
+    hparams = config["data"]
+
+    min_text_len = hparams.get("min_text_len", 1)
+    max_text_len = hparams.get("max_text_len", 190)
+    sampling_rate = hparams.get("sampling_rate", 22050)
+    hop_length = hparams.get("hop_length", 256)
+    print(min_text_len, max_text_len, sampling_rate, hop_length)
+
+    main(in_file, out_file)

wetts/vits/data_utils.py

@@ -4,6 +4,8 @@
 import torch
 import torchaudio
 import torch.utils.data
+from tqdm import tqdm
+import soundfile as sf
 
 from utils.mel_processing import spectrogram_torch, mel_spectrogram_torch
 from utils.task import load_filepaths_and_text
@@ -60,20 +62,26 @@ def _filter(self):
         """
         Filter text & store spec lengths
         """
-        audiopaths_sid_text_new = []
-        lengths = []
-        for item in self.audiopaths_sid_text:
-            audiopath = item[0]
-            src_sampling_rate = torchaudio.info(audiopath).sample_rate
-            # filename|speaker|text
-            text = item[2]
-            if self.min_text_len <= len(text) and len(
-                    text) <= self.max_text_len:
-                audiopaths_sid_text_new.append(item)
-                lengths.append(
-                    int(
-                        os.path.getsize(audiopath) * self.sampling_rate /
-                        src_sampling_rate) // (2 * self.hop_length))
+        if len(self.audiopaths_sid_text[0]) > 3:
+            # spec length is provided
+            audiopaths_sid_text_new = [item[:3] for item in self.audiopaths_sid_text]
+            lengths = [int(item[3]) for item in self.audiopaths_sid_text]
+        else:
+            audiopaths_sid_text_new = []
+            lengths = []
+            for item in tqdm(self.audiopaths_sid_text, desc="Filtering data"):
+                audiopath = item[0]
+                src_sampling_rate = sf.info(audiopath).samplerate
+                # filename|speaker|text
+                text = item[2]
+                text = text.strip().split()
+                if self.min_text_len <= len(text) and len(
+                        text) <= self.max_text_len:
+                    audiopaths_sid_text_new.append(item)
+                    lengths.append(
+                        int(
+                            os.path.getsize(audiopath) * self.sampling_rate /
+                            src_sampling_rate) // (2 * self.hop_length))
         self.audiopaths_sid_text = audiopaths_sid_text_new
         self.lengths = lengths