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

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "f5-tts"
-version = "0.1.1"
+version = "0.1.2"
 description = "F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching"
 readme = "README.md"
 license = {text = "MIT License"}

src/f5_tts/model/dataset.py

@@ -127,38 +127,43 @@ class CustomDataset(Dataset):
         return len(self.data)
 
     def __getitem__(self, index):
-        row = self.data[index]
-        audio_path = row["audio_path"]
-        text = row["text"]
-        duration = row["duration"]
+        while True:
+            row = self.data[index]
+            audio_path = row["audio_path"]
+            text = row["text"]
+            duration = row["duration"]
+
+            # filter by given length
+            if 0.3 <= duration <= 30:
+                break  # valid
+
+            index = (index + 1) % len(self.data)
 
         if self.preprocessed_mel:
             mel_spec = torch.tensor(row["mel_spec"])
-
         else:
             audio, source_sample_rate = torchaudio.load(audio_path)
+
+            # make sure mono input
             if audio.shape[0] > 1:
                 audio = torch.mean(audio, dim=0, keepdim=True)
 
-            if duration > 30 or duration < 0.3:
-                return self.__getitem__((index + 1) % len(self.data))
-
+            # resample if necessary
             if source_sample_rate != self.target_sample_rate:
                 resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)
                 audio = resampler(audio)
 
+            # to mel spectrogram
             mel_spec = self.mel_spectrogram(audio)
-            mel_spec = mel_spec.squeeze(0)  # '1 d t -> d t')
+            mel_spec = mel_spec.squeeze(0)  # '1 d t -> d t'
 
-        return dict(
-            mel_spec=mel_spec,
-            text=text,
-        )
+        return {
+            "mel_spec": mel_spec,
+            "text": text,
+        }
 
 
 # Dynamic Batch Sampler
-
-
 class DynamicBatchSampler(Sampler[list[int]]):
     """Extension of Sampler that will do the following:
     1.  Change the batch size (essentially number of sequences)

src/f5_tts/train/finetune_gradio.py

@@ -1177,7 +1177,10 @@ def get_random_sample_transcribe(project_name):
         sp = item.split("|")
         if len(sp) != 2:
             continue
-        list_data.append([os.path.join(path_project, "wavs", sp[0] + ".wav"), sp[1]])
+
+        # fixed audio when it is absolute
+        file_audio = get_correct_audio_path(sp[0], path_project)
+        list_data.append([file_audio, sp[1]])
 
     if list_data == []:
         return "", None