BANC: Towards Efficient Binaural Audio Neural Codec for Overlapping Speech
Abstract: We introduce BANC, a neural binaural audio codec designed for efficient speech compression in single and two-speaker scenarios while preserving the spatial location information of each speaker. This highly versatile model allows configuration and training tailored to one or two speakers with different spatial locations. Our key contributions are as follows: 1)The ability of our proposed model to compress and decode overlapping speech. 2) A novel architecture that compresses speech content and spatial cues separately, ensuring the preservation of each speaker's spatial context after decoding. 3) BANC's proficiency in reducing the bandwidth required for compressing binaural speech by 48\% compared to compressing individual binaural channels. In our evaluation, we employed speech enhancement, room acoustics, and perceptual metrics to assess the accuracy of BANC's clean speech and spatial cue estimates
Baselines
Audio Demos - Single Speaker
| Audio Codec | Compression | Bandwidth | Reverb Sample-1 | Reverb Sample-2 | Reverb Sample-3 | Reverb Sample-4 |
|---|---|---|---|---|---|---|
| Ground Truth | - | - | ||||
| Opus-12 | - | 12 kbps | ||||
| Opus-24 | - | 24 kbps | ||||
| HiFi-Codec-320 | 320x | - | ||||
| HiFi-Codec-240 | 240x | - | ||||
| Encodec-12 | 256x | 12 kbps | ||||
| Encodec-24 | 64x | 24 kbps | ||||
| AudioDec | 300x | 24 kbps | ||||
| BANC (ours) | 3150x | 12.6 kbps |
Audio Demos - Two Speakers
| Audio Codec | Reverb Sample-1 | Clean Speaker-1 | Clean Speaker-2 | Reverb Sample-2 | Clean Speaker-1 | Clean Speaker-2 | ||
|---|---|---|---|---|---|---|---|---|
| Ground Truth | | | | | ||||||
| Opus-12 | | | | | ||||||
| Encodec-12 | | | | | ||||||
| BANC (ours) | | | | |
Audio Demos - Two Speakers (Training Samples)
| Audio Codec | Training Reverb Sample-1 | Clean Speaker-1 | Clean Speaker-2 | Training Reverb Sample-2 | Clean Speaker-1 | Clean Speaker-2 | ||
|---|---|---|---|---|---|---|---|---|
| Ground Truth | | | | | ||||||
| BANC (ours) | | | | |
Ablation
Audio Demos - Single Speaker
| Audio Codec | Clean-1 | Reverb-1 | Clean-2 | Reverb-2 | Clean-3 | Reverb-3 | Clean-4 | Reverb-4 |
|---|---|---|---|---|---|---|---|---|
| Ground Truth | ||||||||
| BANC-V1 | ||||||||
| BANC-V2 | ||||||||
| BANC (OURS) |
Spetrogram Demos of BANC - Single Speaker
| Reverb Sample | Spetrogram of Ground truth BIR | Spetrogram of estimated BIR |
|---|---|---|
| Reverb Sample-1 |
|
|
| Reverb Sample-2 |
|
|
| Reverb Sample-3 |
|
|
| Reverb Sample-4 |
|
|
Spetrogram Demos of BANC (Reverb Sample-1) - Two Speaker
| Speaker | Spetrogram of Ground truth BIR | Spetrogram of estimated BIR |
|---|---|---|
| Speaker-1 |
|
|
| Speaker-2 |
|
|
Spetrogram Demos of BANC (Training Reverb Sample-1) - Two Speaker
| Speaker | Spetrogram of Ground truth BIR | Spetrogram of estimated BIR |
|---|---|---|
| Speaker-1 |
|
|
| Speaker-2 |
|
|
The box plot for the perceptual evaluation decribed in the paper. We plot the output for each single-speaker reverberant speech signals separately.
| Speaker | Box plot for reverberant speech 1 | Box plot for reverberant speech 2 |
|---|---|---|
| Single Speaker |
|
|
© October 2023
