Text-to-Speech
MLX
ONNX
Portuguese
tts
pt-br
brazilian-portuguese
small-model
quantized
apple-silicon
pingo
fargan
Instructions to use semfence/pingo-ptbr with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use semfence/pingo-ptbr with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download semfence/pingo-ptbr --local-dir pingo-ptbr
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
File size: 1,365 Bytes
27d8618 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 | {
"model_name": "pingo-ptbr",
"weights": "weights/pingo-ptbr.bin",
"acoustic_checkpoint": {"file": "checkpoints/acoustic.pt", "step": 40000, "validation_loss": 0.1699, "width": 64, "ffn": 128, "parameters": 270081},
"vocoder": {"origin": "PiVo-1.4M (Xiph/Opus FARGAN, BSD-3-Clause)", "parameters": 816597, "retrained": false, "opus_revision": "bafd40dcbd506e12da257b8cb05c0afe1ed44719"},
"inference_parameters": 1086678,
"quantization": {
"scheme": "per-layer symmetric per-output-channel, fp16 scales, no calibration, post-training",
"acoustic_bits": {"encoder": 5, "decoder": 6, "durations": 6, "embeddings": 8, "1-D tensors": 32},
"vocoder_bits": "8 by default; 6 for pitch_gain.w, gru1.hidden, glu2, gru1.input, cond.dense2, skip_glu, glu1; 5 for gru3.input, gru2.hidden, glu3, frame_glu, gru3.hidden"
},
"training": {"data": ["semfence/ptbr-synthetic-speech", "semfence/ptbr-sentences"], "train_clips": 19024, "validation_clips": 818, "steps": 40000, "batch": 16, "learning_rate": 3e-4, "gpu": "1x RTX 3090 (rented)", "cost_usd_approx": 0.14},
"voices": {"F_s4": 0, "M_sm4": 1},
"related": {"wide_edition": "weights/pingo-ptbr-w96.bin (pivo-ptbr acoustic model, width 96, quantized only)", "parent": "https://huggingface.co/semfence/pivo-ptbr"},
"onnx_and_browser_build": true,
"listening_test_with_several_people": false
}
|