VoxCPM 1.5¶
Release Date: December 5, 2025
Parameter Size: 750M
Sampling Rate: 44.1kHz
Note
VoxCPM 1.5 is a legacy release for lighter Chinese/English deployments and existing 1.x workflows.
Overview¶
We’re thrilled to introduce a major upgrade that improves audio quality and efficiency of VoxCPM, while maintaining the core capabilities of context-aware speech generation and zero-shot voice cloning.
Feature |
VoxCPM |
VoxCPM1.5 |
|---|---|---|
Audio VAE Sampling Rate |
16kHz |
44.1kHz |
LM Token Rate |
12.5Hz |
6.25Hz |
Patch Size |
2 |
4 |
Token Rate |
12.5Hz |
6.25Hz |
Getting Started¶
For installation, loading, and the shared generation API, start with Quick Start.
Choose VoxCPM 1.5 when you want a lighter Chinese/English checkpoint than VoxCPM 2 while keeping higher output quality than VoxCPM 1.0.
Model Updates¶
AudioVAE Sampling Rate: 16kHz -> 44.1kHz¶
The AudioVAE now supports 44.1kHz sampling rate, which allows the model to:
Better cloning fidelity with more preserved high-frequency detail
Note
This upgrade enables higher quality generation when using high-quality reference audio, but does not guarantee that all generated audio will be high-fidelity. The output quality depends on the prompt speech quality.
Token Rate: 12.5Hz -> 6.25Hz¶
We reduced the token rate in LM from 12.5Hz to 6.25Hz (patch size increased from 2 to 4) while maintaining similar performance on evaluation benchmarks. This change:
Reduces computational requirements for generating the same length of audio
Provides a foundation for longer audio generation
Migration Guide¶
From VoxCPM-0.5B to VoxCPM1.5¶
Update Model Path: Point to VoxCPM1.5 checkpoint
Update Sample Rate: Change
sample_ratefrom16000to44100in configsUpdate Audio Data: The current fine-tuning pipeline automatically resamples to 44.1kHz. If you preprocess offline, resample to 44.1kHz.
Review Training Parameters: Adjust batch size if needed due to higher sampling rate
Backward Compatibility¶
VoxCPM-0.5B models and configurations remain fully supported
Code automatically detects and adapts to model version
No breaking changes to the API