We propose TQCodec, a neural audio codec designed for high-bitrate, high-fidelity music streaming. Unlike existing neural codecs that primarily target ultra-low bitrates (<= 16kbps), TQCodec operates at 44.1 kHz and supports bitrates from 32 kbps to 128 kbps, aligning with the standard quality of modern music streaming platforms. The model adopts an encoder-decoder architecture based on SEANet for efficient on-device computation and introduces several enhancements: an imbalanced network design for improved quality with low overhead, SimVQ for mid-frequency detail preservation, and a phase-aware waveform loss. Additionally, we introduce a perception-driven band-wise bit allocation strategy to prioritize perceptually critical lower frequencies. Evaluations on diverse music datasets demonstrate that TQCodec achieves superior audio quality at target bitrates, making it well-suited for high-quality audio applications.
Figures & tables
Codec
SR (khz)
BR (kbps)
CB
FR
TQCodec
44.1
32, 64, 128
5/ 10/ 20
689
DAC [ 6 ]
44.1
16
18
86
Encodec [ 3 ]
24/ 48
24
16
75 /150
Table 1: Comparison of compression approaches (SR: sample rate, BR: bit rate, CB: codebook, FR: frame rate).
Figure 1: Illustration of our subband modeling.
Name
Number of tracks
MusDBHQ [ 12 ]
150
Jingju [ 4 ]
120
Jamendo [ 1 ]
55,609
Fma [ 2 ]
106,574
Private dataset
100000+
Table 2: The dataset information.
Figure 2: Spectrogram visualization, where the RVQ fails to reconstruct the mid-frequency clearly (above 4000Hz).
Name
LSD ↓
LSD-L ↓
LSD-H ↓
MACs-E ↓
MACs-D ↓
Receptive field ↓
DAC (16kbps)
0.927
0.877
1.016
25G
365G
17706
DAC-small (32kbps)
0.833
0.781
0.936
25G
365G
17706
Ablation on TQCodec (32kbps)
Encodec + DAC RVQ
0.854
0.790
0.983
2.4G
6.31G
2410
+ RSimVQ
0.850
0.772
0.997
2.4G
6.31G
2410
+ Waveform loss
0.844
0.769
0.995
2.4G
6.31G
2410
Table 3: Ablation study results of the proposed codec, where LSD-L refers to frequency below 16000Hz and LSD-H refers to frequency above 16000Hz.
Model
Bitrates
LSD-L ↓
LSD-H ↓
SNR ↑
Ogg-Vorbis [ 11 ]
48kbps
0.763
2.847
16.998
TQCodec
43kbps
0.718
0.954
16.074
+ Subband
43kbps
0.702
1.163
16.88
Table 4: Comparison with baselines to verify the subband modeling.
MOS
Average
4.18
Pass Rate
100%
Excellent Rate
77.8%
Preference
HQ (Ogg) Better
39.4%
AICodec Better
39.1%
Table 5: Subjective Listening Test: AICodec-64kbps vs. Ogg-96kbps (HQ)
Indiana University, Bloomington, IN, USA · Electronics and Telecommunications Research Institute, Daejeon, South Korea · University of Illinois at Urbana-Champaign, Urbana, IL, USA