Upload 3 files

app.py

@@ -0,0 +1,31 @@
+import gradio as gr
+import soundfile as sf
+import tempfile
+from f5_tts_loader import F5TTS
+
+tts = F5TTS("hynt/F5-TTS-Vietnamese-100h")
+
+
+def generate_speech(text):
+    if not text.strip():
+        return None
+
+    audio, sr = tts.tts(text)
+
+    # Save to temp WAV
+    out_path = tempfile.mktemp(suffix=".wav")
+    sf.write(out_path, audio, sr)
+
+    return out_path
+
+
+with gr.Blocks(title="Vietnamese TTS Free") as demo:
+    gr.Markdown("# 🇻🇳 Vietnamese Text-to-Speech (F5-TTS, Free)\nNhập tiếng Việt để tạo giọng nói:")
+
+    text_input = gr.Textbox(lines=5, label="Văn bản")
+    audio_output = gr.Audio(label="Âm thanh", type="filepath")
+    btn = gr.Button("🎤 Convert")
+
+    btn.click(fn=generate_speech, inputs=text_input, outputs=audio_output)
+
+demo.launch()

f5_tts_loader.py

@@ -0,0 +1,32 @@
+import torch
+import torchaudio
+from transformers import AutoTokenizer
+
+
+class F5TTS:
+    def __init__(self, model_name="hynt/F5-TTS-Vietnamese-100h"):
+        self.device = torch.device("cpu")
+
+        # Load tokenizer
+        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+        # Load model weights
+        self.model = torch.load(
+            self._download_model(model_name),
+            map_location=self.device
+        )
+
+        self.model.eval()
+
+    def _download_model(self, repo):
+        """Download model file from HuggingFace repo."""
+        from huggingface_hub import hf_hub_download
+        return hf_hub_download(repo_id=repo, filename="model.safetensors")
+
+    def tts(self, text, sample_rate=22050):
+        tokens = self.tokenizer(text, return_tensors="pt")["input_ids"].to(self.device)
+
+        with torch.no_grad():
+            audio = self.model.generate(tokens)[0].cpu().numpy()
+
+        return audio, sample_rate

requirements.txt

@@ -0,0 +1,9 @@
+torch
+torchaudio
+gradio
+numpy
+soundfile
+einops
+transformers
+accelerate
+sentencepiece