Sync from GitHub repo

This Space is synced from the GitHub repo: https://github.com/SWivid/F5-TTS. Please submit contributions to the Space there

src/f5_tts/infer/speech_edit.py

@@ -9,6 +9,7 @@ import torch.nn.functional as F
 import torchaudio
 from hydra.utils import get_class
 from omegaconf import OmegaConf
+from cached_path import cached_path
 
 from f5_tts.infer.utils_infer import load_checkpoint, load_vocoder, save_spectrogram
 from f5_tts.model import CFM
@@ -55,7 +56,8 @@ win_length = model_cfg.model.mel_spec.win_length
 n_fft = model_cfg.model.mel_spec.n_fft
 
 
-ckpt_path = str(files("f5_tts").joinpath("../../")) + f"ckpts/{exp_name}/model_{ckpt_step}.safetensors"
+# ckpt_path = str(files("f5_tts").joinpath("../../")) + f"/ckpts/{exp_name}/model_{ckpt_step}.safetensors"
+ckpt_path = str(cached_path(f"hf://SWivid/F5-TTS/{exp_name}/model_{ckpt_step}.safetensors"))
 output_dir = "tests"
 
 

src/f5_tts/train/datasets/prepare_emilia_v2.py

@@ -0,0 +1,94 @@
+# put in src/f5_tts/train/datasets/prepare_emilia_v2.py
+# prepares Emilia dataset with the new format w/ Emilia-YODAS
+
+import os
+import json
+from concurrent.futures import ProcessPoolExecutor
+from pathlib import Path
+from tqdm import tqdm
+from datasets.arrow_writer import ArrowWriter
+from importlib.resources import files
+
+from f5_tts.model.utils import (
+    repetition_found,
+)
+
+# Define filters for exclusion
+out_en = set()
+en_filters = ["ا", "い", "て"]
+
+
+def process_audio_directory(audio_dir):
+    sub_result, durations, vocab_set = [], [], set()
+    bad_case_en = 0
+
+    for file in audio_dir.iterdir():
+        if file.suffix == ".json":
+            with open(file, "r") as f:
+                obj = json.load(f)
+                text = obj["text"]
+                if any(f in text for f in en_filters) or repetition_found(text, length=4):
+                    bad_case_en += 1
+                    continue
+
+                duration = obj["duration"]
+                audio_file = file.with_suffix(".mp3")
+                if audio_file.exists():
+                    sub_result.append({"audio_path": str(audio_file), "text": text, "duration": duration})
+                    durations.append(duration)
+                    vocab_set.update(list(text))
+
+    return sub_result, durations, vocab_set, bad_case_en
+
+
+def main():
+    assert tokenizer in ["pinyin", "char"]
+    result, duration_list, text_vocab_set = [], [], set()
+    total_bad_case_en = 0
+
+    executor = ProcessPoolExecutor(max_workers=max_workers)
+    futures = []
+    dataset_path = Path(dataset_dir)
+    for sub_dir in dataset_path.iterdir():
+        if sub_dir.is_dir():
+            futures.append(executor.submit(process_audio_directory, sub_dir))
+
+    for future in tqdm(futures, total=len(futures)):
+        sub_result, durations, vocab_set, bad_case_en = future.result()
+        result.extend(sub_result)
+        duration_list.extend(durations)
+        text_vocab_set.update(vocab_set)
+        total_bad_case_en += bad_case_en
+
+    executor.shutdown()
+
+    if not os.path.exists(f"{save_dir}"):
+        os.makedirs(f"{save_dir}")
+
+    with ArrowWriter(path=f"{save_dir}/raw.arrow") as writer:
+        for line in tqdm(result, desc="Writing to raw.arrow ..."):
+            writer.write(line)
+
+    with open(f"{save_dir}/duration.json", "w", encoding="utf-8") as f:
+        json.dump({"duration": duration_list}, f, ensure_ascii=False)
+
+    with open(f"{save_dir}/vocab.txt", "w") as f:
+        for vocab in sorted(text_vocab_set):
+            f.write(vocab + "\n")
+
+    print(f"For {dataset_name}, sample count: {len(result)}")
+    print(f"For {dataset_name}, vocab size is: {len(text_vocab_set)}")
+    print(f"For {dataset_name}, total {sum(duration_list) / 3600:.2f} hours")
+    print(f"Bad en transcription case: {total_bad_case_en}\n")
+
+
+if __name__ == "__main__":
+    max_workers = 32
+    tokenizer = "char"
+    dataset_dir = "/home/ubuntu/emilia-dataset/Emilia-YODAS/EN"
+    dataset_name = f"Emilia_EN_{tokenizer}"
+    # save_dir = os.path.expanduser(f"~/F5-TTS/data/{dataset_name}")
+    save_dir = str(files("f5_tts").joinpath("../../")) + f"/data/{dataset_name}"
+
+    print(f"Prepare for {dataset_name}, will save to {save_dir}\n")
+    main()