Implementation:Speechbrain Speechbrain Prepare Voicebank CTC
| Knowledge Sources | |
|---|---|
| Domains | Speech_Enhancement, Data_Preparation |
| Last Updated | 2026-02-09 00:00 GMT |
Overview
Concrete tool for preparing the Voicebank/VCTK dataset for ASR and speech enhancement tasks provided by the SpeechBrain library.
Description
This script prepares JSON manifest files for the Voicebank (VCTK) dataset, handling both clean and noisy speech recordings. It processes audio at 16kHz, splits speakers into train/valid/test sets, generates phoneme-level lexicon entries from the CMU dictionary, and creates JSON manifests containing paths to clean audio, noisy audio, and text transcriptions. The script also includes a download utility for obtaining and resampling the VCTK dataset.
Usage
Use this script when preparing data for CTC-based ASR or speech enhancement experiments on the Voicebank/VCTK corpus. It should be executed before running any Voicebank training recipe.
Code Reference
Source Location
- Repository: SpeechBrain
- File: recipes/Voicebank/ASR/CTC/voicebank_prepare.py
Signature
def prepare_voicebank(
data_folder, save_folder, valid_speaker_count=2, skip_prep=False
):
Import
from voicebank_prepare import prepare_voicebank
I/O Contract
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
| data_folder | str | Yes | Path to the folder where the original Voicebank dataset is stored |
| save_folder | str | Yes | The directory where to store the JSON files |
| valid_speaker_count | int | No | Number of validation speakers out of 28 in the train set (default: 2) |
| skip_prep | bool | No | If True, skips data preparation (default: False) |
Outputs
| Name | Type | Description |
|---|---|---|
| train.json | JSON file | Training manifest with clean/noisy audio paths and transcriptions |
| valid.json | JSON file | Validation manifest |
| test.json | JSON file | Test manifest |
Usage Examples
from voicebank_prepare import prepare_voicebank
prepare_voicebank(
data_folder="/path/to/datasets/Voicebank",
save_folder="exp/Voicebank_exp",
)
# With custom validation speaker count
prepare_voicebank(
data_folder="/path/to/datasets/Voicebank",
save_folder="exp/Voicebank_exp",
valid_speaker_count=4,
)