Zum Inhalt springen
ModelAufgelistetExtern
Herunterladen

Neuphonic Neucodec Onnx Decoder Int8 Mirror

This is an onnx-compiled version of the decoder of NeuCodec, quantized to int8. It's main use case is providing a low footprint decoder for on-device TTS.

ArtOtherBasismodellneuphonic/neucodecVersionRevision e643cf1CBewertung

Modellquelle

Art
Other
Dateigröße
298 MB
Basismodell
neuphonic/neucodec
Version
vRevision e643cf1
Aktualisiert
31. Okt. 2026
Downloads
52
Quelle
Hugging Face

Quellenbeschreibung

NeuCodec Onnx Decoder

This is an onnx-compiled version of the decoder of NeuCodec, quantized to int8. It's main use case is providing a low footprint decoder for on-device TTS.

Usage

import librosa
import torch
import soundfile as sf
from torchaudio import transforms as T
from neucodec import NeuCodec, NeuCodecOnnxDecoder
 
model = NeuCodec.from_pretrained("neuphonic/neucodec")
model.eval()
compiled_model = NeuCodecOnnxDecoder.from_pretrained("neuphonic/neucodec-onnx-decoder-int8")
 
y, sr = torchaudio.load(librosa.ex("libri1"))
if sr != 16_000:
    y = T.Resample(sr, 16_000)(y)[None, ...] # (B, 1, T_16)

with torch.no_grad():
    fsq_codes = model.encode_code(y)
    # fsq_codes = model.encode_code(librosa.ex("libri1")) # or directly pass your filepath!
    print(f"Codes shape: {fsq_codes.shape}")  
    recon = compiled_model.decode_code(fsq_codes) # (B, 1, T_24)

sf.write("reconstructed.wav", recon, 24_000)

Quellen

1 Quelle
Hugging FacePrimärVersion e643cf108cdb80c41e5741bf212a785d115eb79c
huggingface.co/aoiandroid/neuphonic-neucodec-onnx-decoder-int8-mirror/tree/e643cf108cdb80c41e5741bf212a785d115eb79c

Verifiziert 31. Okt.

Modellartefakte

1 Artefakt

model.onnx

onnx · 298 MB · SHA-256 3ddd9e56396e…a6b3 · Hugging Face

Download file

Quellenauszüge

3 Auszüge

--- license: apache-2.0 base_model: - neuphonic/neucodec pipeline_tag: audio-to-audio --- # NeuCodec Onnx Decoder This is an onnx-compiled version of the decoder of [NeuCodec](https://huggingface.co/neuphonic/neucodec), quantized to int8. It's main use case is providing a low footprint decoder for on-device TTS. ## Usage ```python import librosa import torch import soundfile as sf from torchaudio import transforms as T from neucodec import NeuCodec, NeuCodecOnnxDecoder model = NeuCodec.from_pretrained("neuphonic/neucodec") model.eval() compiled_model = NeuCodecOnnxDecoder.from_pretrained("neuphonic/neucodec-onnx-decoder-int8") y, sr = torchaudio.load(librosa.ex("libri1")) if sr != 16_000: y = T.Resample(sr, 16_000)(y)[None, ...] # (B, 1, T_16) with torch.no_grad(): fsq_codes = model.encode_code(y) # fsq_codes = model.encode_code(librosa.ex("libri1")) # or directly pass your filepath! print(f"Codes shape: {fsq_codes.shape}") recon = compiled_model.decode_code(fsq_codes) # (B, 1, T_24) sf.write("reconstructed.wav", recon, 24_000) ```

Source context: 52 downloads · 0 likes · Pipeline audio-to-audio · Repo aoiandroid/neuphonic-neucodec-onnx-decoder-int8-mirror

Source context: 1 downloads · 0 likes · Pipeline audio-to-audio · Repo aoiandroid/neuphonic-neucodec-onnx-decoder-int8-mirror