OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Audio & Voice AI

Sincnet

SincNet is a Python and PyTorch convolutional neural network architecture for raw-audio processing, with a first layer of parametrized sinc-function band-pass filters whose cutoff frequencies are learned from data.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Sincnet — OpenViST project profile in Audio & Voice AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use SincNet when you want to study or build a raw-audio neural model with a compact, learnable band-pass filter bank, or reproduce the repository’s TIMIT speaker-identification workflow.

Who it is for

  • Researchers studying neural architectures for raw audio and speech.
  • Developers building speaker-identification experiments or audio-processing pipelines.

Documented uses

  • Build a raw-audio classification model using a parametrized first convolutional layer.
  • Prepare the TIMIT corpus and train the provided speaker-identification example.
  • Explore a learnable filter-bank front end for speech-recognition experiments.
Before you choose

Requirements & limits

The repository is public and MIT-licensed.

  • The first convolutional layer learns low and high frequency cutoffs rather than every filter element.
  • The repository includes a TIMIT speaker-identification example and code for preparing normalized speech data.
  • The documented setup specifies Linux, Python 3.6 or 2.7, PyTorch 1.0 and pysoundfile.
  • The README identifies PyTorch-Kaldi as a place to explore SincNet for speech-recognition experiments.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare TIMIT and run the speaker-identification example

Prepare a normalized TIMIT dataset, configure the supplied experiment and start training the speaker-identification model.

Before you start

  • Linux, Python 3.6 or 2.7, PyTorch 1.0 and pysoundfile.
  • The original TIMIT corpus and the repository’s data_lists/TIMIT_all.scp file.

Steps

  1. Run the command below, substituting the original corpus path and the desired output path.

    Documented command
    python TIMIT_preparation.py $TIMIT_FOLDER $OUTPUT_FOLDER data_lists/TIMIT_all.scp

  2. Edit the [data] section of cfg/SincNet_TIMIT.cfg, setting data_folder to the prepared dataset location.

  3. Run the command below to start the experiment.

    Documented command
    python speaker_id.py --cfg=cfg/SincNet_TIMIT.cfg

Expected outcome

The README says results are saved in the configured output folder, including training and test error summaries in res.res and a model in model_raw.pkl.

Guide limitations

  • Training may take several hours; the README reports about 24 hours for its example on an NVIDIA TITAN X.
  • The code frequently accesses stored WAV files; the README advises copying the normalized dataset to the local node when running on a cluster.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
mravanelli/SincNet · Official homepage title · retained in its source language
How we review sourcesSuggest a correction

Related options

SpeechBrain — The README says SincNet is also implemented in SpeechBrain, a PyTorch-based speech-processing toolkit.

Explore project →

PyTorch-Kaldi — The README points to this project for SincNet applied to speech-recognition experiments.