Implementation:Speechbrain Speechbrain Create Aishell1Mix From Metadata

Knowledge Sources	SpeechBrain
Domains	Speech Separation, Data Preparation
Last Updated	2026-02-09 00:00 GMT

Overview

Concrete tool for creating Aishell1Mix audio mixtures from pre-generated metadata provided by the SpeechBrain library.

Description

This script reads metadata CSV files generated by the companion metadata creation script and produces the actual mixed audio files for the Aishell1Mix speech separation dataset. It combines source utterances from AISHELL-1 with optional WHAM! noise at specified frequencies (8kHz, 16kHz), modes (min, max), and mixture types (mix_clean, mix_both, mix_single). The script supports parallel processing for efficient generation of large-scale mixture datasets.

Usage

Use this when generating the Aishell1Mix mixed audio files from previously created metadata for speech separation training with SpeechBrain recipes.

Code Reference

Source Location

• Repository: SpeechBrain
• File: recipes/Aishell1Mix/separation/scripts/create_aishell1mix_from_metadata.py

Signature

def create_aishell1mix(
    aishell1_dir, wham_dir, out_dir, metadata_dir, freqs, n_src, modes, types
):

CLI Interface

python create_aishell1mix_from_metadata.py \
    --aishell1_dir /path/to/aishell1/data_aishell/wav \
    --wham_dir /path/to/wham_noise \
    --metadata_dir /path/to/aishell1mix/metadata \
    --aishell1mix_outdir /path/to/aishell1mix \
    --n_src 2 \
    --freqs 8k 16k \
    --modes min max \
    --types mix_clean mix_both mix_single

I/O Contract

Inputs

Name	Type	Required	Description
aishell1_dir	str	Yes	Path to AISHELL-1 root directory containing WAV files
wham_dir	str	Yes	Path to WHAM! noise root directory
metadata_dir	str	Yes	Path to the Aishell1Mix metadata directory with CSV files
aishell1mix_outdir	str	Yes	Path to the desired output dataset root directory
n_src	int	No	Number of sources in mixtures (default: 2)
freqs	list[str]	No	Target frequencies, e.g., ["8k", "16k"] (default: ["8k", "16k"])
modes	list[str]	No	Mixing modes, e.g., ["min", "max"] (default: ["min", "max"])
types	list[str]	No	Mixture types (default: ["mix_clean", "mix_both", "mix_single"])

Outputs

Name	Type	Description
Mixed WAV files	WAV Audio Files	Generated mixture and source audio files organized by frequency, mode, and type

Usage Examples

# Typically invoked via command line:
# python create_aishell1mix_from_metadata.py --aishell1_dir /data/aishell1/wav --wham_dir /data/wham_noise --metadata_dir /data/aishell1mix/metadata --aishell1mix_outdir /data/aishell1mix

# Or programmatically:
from create_aishell1mix_from_metadata import create_aishell1mix

create_aishell1mix(
    aishell1_dir="/data/aishell1/data_aishell/wav",
    wham_dir="/data/wham_noise",
    out_dir="/data/Aishell1Mix2",
    metadata_dir="/data/aishell1mix/metadata",
    freqs=["8k", "16k"],
    n_src=2,
    modes=["min", "max"],
    types=["mix_clean", "mix_both", "mix_single"],
)

Related Pages

• Implementation:Speechbrain_Speechbrain_Create_Aishell1Mix_Metadata -- Companion script for generating metadata
• Principle:Speechbrain_Speechbrain_Dataset_Specific_Data_Preparation