F5-TTS Hindi

Fine-tuned F5-TTS model for Hindi text-to-speech.

Training Details

• Base model: SPRINGLab/F5-Hindi-24KHz (151M params, F5-TTS Small)
• Datasets: IndicTTS-Hindi + IndicVoices-R_Hindi
• Learning rate: 1e-05
• Epochs: 10
• Batch size: 200 frames/GPU
• Architecture: DiT (dim=768, depth=18, heads=12) + ConvNeXt V2 (dim=512, layers=4)
• Audio: 24kHz, 100-dim mel spectrogram, Vocos vocoder

Usage

from f5_tts.api import F5TTS
model = F5TTS(
 model="F5-TTS",
 ckpt_file="path/to/model_last.pt", 
 vocab_file="path/to/vocab.txt",
)
wav, sr, _ = model.infer(
 ref_file="reference.wav",
 ref_text="संदर्भ पाठ हिंदी में",
 gen_text="आपका दिन शुभ हो",
)

training loss

👁 Screenshot 2026-06-01 at 7.05.47 AM

Base model