Bi²: Phase-Aware Binary Mask in Bispectral Domain for Single-Channel Speech Enhancement

Siarhei Y. Barysenka, Pejman Mowlaee

Living Room, SNR = 2.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram

Living Room, SNR = 7.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram

Office, SNR = 2.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram

Office, SNR = 7.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram

Cafe, SNR = 2.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram

Cafe, SNR = 7.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram

Public Square, SNR = 2.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram

Public Square, SNR = 7.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram

Bus, SNR = 2.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram

Bus, SNR = 7.5 dB

Noisy

Noisy speech spectrogram Noisy speech spectrogram

Bi²

Bi² enhanced speech spectrogram Bi² enhanced speech spectrogram

L.K.

L.K. enhanced speech spectrogram L.K. enhanced speech spectrogram

ijMAP

ijMAP enhanced speech spectrogram ijMAP enhanced speech spectrogram

DTLN

DTLN enhanced speech spectrogram DTLN enhanced speech spectrogram

L.K. (Oracle)

L.K. oracle enhanced speech spectrogram L.K. oracle enhanced speech spectrogram

Clean

Clean speech spectrogram Clean speech spectrogram